Email alert
2024 Vol. 46, No. 3
Display Method:
2024, 46(3): 1-4.
Abstract:
2024, 46(3): 765-776.
doi: 10.11999/JEIT230390
Abstract:
With the rapid increase in the number of network access devices and the volume of network access data currently, the shortcomings of the centralized computing architecture represented by cloud computing are increasingly exposed. Edge computing, that is making computing as close to the data source as possible to reduce data transmission time and network delay, has become the focus of academia and industry as a supplement to cloud computing. An instance architecture widely used in edge computing: Cloud-Edge-Terminal architecture, and a typical application of edge computing: edge intelligent computing is focused on in this paper. Two key issues of edge intelligent computing under Cloud-Edge-Terminal architecture: computing optimization and computing offloading is analyzed. First, the research focus of edge intelligent computing is analyzed, and the application and research status of intelligent computing optimization under Cloud-Edge-Terminal architecture is combed. Then the research ideas and current situation of computing offloading under Cloud-Edge-Terminal architecture is discussed. Finally, the challenges and research trends of edge intelligent computing under Cloud-Edge-Terminal architecture is summarized.
With the rapid increase in the number of network access devices and the volume of network access data currently, the shortcomings of the centralized computing architecture represented by cloud computing are increasingly exposed. Edge computing, that is making computing as close to the data source as possible to reduce data transmission time and network delay, has become the focus of academia and industry as a supplement to cloud computing. An instance architecture widely used in edge computing: Cloud-Edge-Terminal architecture, and a typical application of edge computing: edge intelligent computing is focused on in this paper. Two key issues of edge intelligent computing under Cloud-Edge-Terminal architecture: computing optimization and computing offloading is analyzed. First, the research focus of edge intelligent computing is analyzed, and the application and research status of intelligent computing optimization under Cloud-Edge-Terminal architecture is combed. Then the research ideas and current situation of computing offloading under Cloud-Edge-Terminal architecture is discussed. Finally, the challenges and research trends of edge intelligent computing under Cloud-Edge-Terminal architecture is summarized.
2024, 46(3): 777-789.
doi: 10.11999/JEIT230357
Abstract:
Traffic sign detection plays an important role in the safe and stable operation of intelligent transportation systems and intelligent driving. Unbalanced data distribution and monotonous scene will lead to poor model performance, but building a complete real traffic scene dataset requires expensive time and labor costs. Based on this, a new metaverse-oriented traffic sign detection paradigm is proposed to alleviate the dependence of existing methods on real data. Firstly, by establishing the scene mapping and model mapping between the metaverse and the physical world, the efficient operation of the detection algorithm between the virtual and real worlds is realized. As a virtualized digital world, Metaverse can complete custom scene construction based on the physical world, and provide massive and diverse virtual scene data for the model. At the same time, knowledge distillation and the mean teacher model is combined in this paper to establish a model mapping to deal with the problem of data differences between the metaverse and the physical world. Secondly, in order to further improve the adaptability of the training model under the Metaverse to the real driving environment, a heuristic attention mechanism is designed to improve the generalization ability of the detection model by locating and learning features. The proposed architecture is experimentally verified on the CURE-TSD, KITTI, Virtual KITTI (VKITTI) datasets. Experimental results show that the proposed metaverse-oriented traffic sign detector has excellent detection results in the physical world without relying on a large number of real scenes, and the detection accuracy reaches 89.7%, which is higher than other detection methods of recent years.
Traffic sign detection plays an important role in the safe and stable operation of intelligent transportation systems and intelligent driving. Unbalanced data distribution and monotonous scene will lead to poor model performance, but building a complete real traffic scene dataset requires expensive time and labor costs. Based on this, a new metaverse-oriented traffic sign detection paradigm is proposed to alleviate the dependence of existing methods on real data. Firstly, by establishing the scene mapping and model mapping between the metaverse and the physical world, the efficient operation of the detection algorithm between the virtual and real worlds is realized. As a virtualized digital world, Metaverse can complete custom scene construction based on the physical world, and provide massive and diverse virtual scene data for the model. At the same time, knowledge distillation and the mean teacher model is combined in this paper to establish a model mapping to deal with the problem of data differences between the metaverse and the physical world. Secondly, in order to further improve the adaptability of the training model under the Metaverse to the real driving environment, a heuristic attention mechanism is designed to improve the generalization ability of the detection model by locating and learning features. The proposed architecture is experimentally verified on the CURE-TSD, KITTI, Virtual KITTI (VKITTI) datasets. Experimental results show that the proposed metaverse-oriented traffic sign detector has excellent detection results in the physical world without relying on a large number of real scenes, and the detection accuracy reaches 89.7%, which is higher than other detection methods of recent years.
2024, 46(3): 790-797.
doi: 10.11999/JEIT230257
Abstract:
Brain-Computer Interface (BCI) can convert the brain activity related to the subject's intention into external device control instructions, which have high application potential in treating neurological diseases, motor rehabilitation, and other aspects. Considering that the materialization of BCI needs to obtain meaningful signals from the human brain, ElectroEncephaloGraphy (EEG) and Near-InfRared Spectroscopy (NIRS) has become important signal acquisition methods for BCI because they are non-invasive, convenient to wear, and relatively cheap. EEG reflects neural electrical activity and is widely applied in BCI systems with high real-time response requirements; NIRS mainly reflects the level of hemodynamics and is mainly utilized in research with precise localization of active brain regions, such as identifying neurophysiological status. Compared with the single-mode BCI system, the BCI system based on EEG-NIRS combined analysis has attracted interest and research in physiological state detection, motor imagination, etc., because of its richer signal characteristics. This review begins with the application of EEG-NIRS combined data analysis in BCI, summarizes the current development on the data and feature fusion level, and looks forward to the research prospects of EEG-NIRS signal processing methods.
Brain-Computer Interface (BCI) can convert the brain activity related to the subject's intention into external device control instructions, which have high application potential in treating neurological diseases, motor rehabilitation, and other aspects. Considering that the materialization of BCI needs to obtain meaningful signals from the human brain, ElectroEncephaloGraphy (EEG) and Near-InfRared Spectroscopy (NIRS) has become important signal acquisition methods for BCI because they are non-invasive, convenient to wear, and relatively cheap. EEG reflects neural electrical activity and is widely applied in BCI systems with high real-time response requirements; NIRS mainly reflects the level of hemodynamics and is mainly utilized in research with precise localization of active brain regions, such as identifying neurophysiological status. Compared with the single-mode BCI system, the BCI system based on EEG-NIRS combined analysis has attracted interest and research in physiological state detection, motor imagination, etc., because of its richer signal characteristics. This review begins with the application of EEG-NIRS combined data analysis in BCI, summarizes the current development on the data and feature fusion level, and looks forward to the research prospects of EEG-NIRS signal processing methods.
2024, 46(3): 798-807.
doi: 10.11999/JEIT230296
Abstract:
Considering the multi-user Non-Orthogonal Multi-Access(NOMA) system under the full-duplex attack of passive eaves-dropping and active interference at the same time, a robust beamforming scheme based on Reconfigurable Intelligent Surfaces (RIS) is proposed to realize physical layer secure communication. Under the condition that only the channel state information of the eavesdropper is known, the system transmission interruption probability and the confidential interruption probability are taken as the constraints and the system confidentiality rate is maximized by jointly optimizing the base station transmission beamforming, RIS phase shift matrix, transmission rate and redundancy rate. In order to solve the above multivariate coupling non-convex optimization problem, an effective alternating optimization algorithm is proposed to obtain the suboptimal solution of the joint optimization problem. The simulation results show that the proposed scheme can achieve a higher confidentiality rate, and when the number of RIS reflective elements is increased, the system confidentiality performance gets better.
Considering the multi-user Non-Orthogonal Multi-Access(NOMA) system under the full-duplex attack of passive eaves-dropping and active interference at the same time, a robust beamforming scheme based on Reconfigurable Intelligent Surfaces (RIS) is proposed to realize physical layer secure communication. Under the condition that only the channel state information of the eavesdropper is known, the system transmission interruption probability and the confidential interruption probability are taken as the constraints and the system confidentiality rate is maximized by jointly optimizing the base station transmission beamforming, RIS phase shift matrix, transmission rate and redundancy rate. In order to solve the above multivariate coupling non-convex optimization problem, an effective alternating optimization algorithm is proposed to obtain the suboptimal solution of the joint optimization problem. The simulation results show that the proposed scheme can achieve a higher confidentiality rate, and when the number of RIS reflective elements is increased, the system confidentiality performance gets better.
2024, 46(3): 808-816.
doi: 10.11999/JEIT230160
Abstract:
Terahertz communication, as one of the key technologies for 6G, is considered an effective means of addressing the scarcity of spectrum resources and improving system capacity. However, due to high path loss and the molecule absorption, terahertz is easily blocked by obstacles leading to communication interruptions. To address this problem, Reconfigurable Intelligent Surface (RIS) is introduced into terahertz communication systems and the impact of channel uncertainty on transmission stability is considered to establish a multi-user energy-efficiency maximization beamforming model based on user quality of service constraints, base station transmit power constraints and RIS discrete phase shift constraints. The original nonconvex optimization problem is solved by transforming it into a convex optimization problem using Dinkelbach, continuous convex approximation, S-procedure, semi-positive definite relaxation, phase mapping and block coordinate descent. Simulation results show that the proposed algorithm improves the energy efficiency by 15.4% and reduces the outage probability by 15.48% compared with the traditional non-robust beamforming algorithm.
Terahertz communication, as one of the key technologies for 6G, is considered an effective means of addressing the scarcity of spectrum resources and improving system capacity. However, due to high path loss and the molecule absorption, terahertz is easily blocked by obstacles leading to communication interruptions. To address this problem, Reconfigurable Intelligent Surface (RIS) is introduced into terahertz communication systems and the impact of channel uncertainty on transmission stability is considered to establish a multi-user energy-efficiency maximization beamforming model based on user quality of service constraints, base station transmit power constraints and RIS discrete phase shift constraints. The original nonconvex optimization problem is solved by transforming it into a convex optimization problem using Dinkelbach, continuous convex approximation, S-procedure, semi-positive definite relaxation, phase mapping and block coordinate descent. Simulation results show that the proposed algorithm improves the energy efficiency by 15.4% and reduces the outage probability by 15.48% compared with the traditional non-robust beamforming algorithm.
2024, 46(3): 817-825.
doi: 10.11999/JEIT230251
Abstract:
A high accuracy combined positioning method for high-speed trains based on the Improved Adaptive Interacting Multiple Model (IMM) is proposed for the high-precision positioning problem of trains. Firstly, a combined positioning scheme of four sensors, namely, satellite receiver, wheel speed sensor, speed radar and single-axis gyroscope, is designed according to the train positioning requirements and the characteristics of each sensor. Next, to address the issue that the IMM fusion filtering algorithm has improper fixed parameter settings due to inaccurate a priori information, the Sage-Husa adaptive filtering and the Transition Probability Matrix (TPM) adaptive update set are introduced to become the adaptive IMM algorithm. To solve the lag problem of multi-model switching, the likelihood function value is set as the judgment flag by using the feature that sub-model likelihood function value can quickly respond to the model change trend, and the judgment window is introduced to correct the TPM matrix elements, which effectively improves the model switching speed. Finally, based on the improved adaptive IMM algorithm, the fusion filtering of four sensor positioning information is carried out to realize the high-precision combined positioning of high-speed trains. Simulation results show that the enhanced algorithm improves the positioning accuracy by 1.6%～14.7% compared with other adaptive IMM algorithms, and it can effectively reduce the peak positional error by increasing the switching speed between models, and it also has a better anti-noise performance.
A high accuracy combined positioning method for high-speed trains based on the Improved Adaptive Interacting Multiple Model (IMM) is proposed for the high-precision positioning problem of trains. Firstly, a combined positioning scheme of four sensors, namely, satellite receiver, wheel speed sensor, speed radar and single-axis gyroscope, is designed according to the train positioning requirements and the characteristics of each sensor. Next, to address the issue that the IMM fusion filtering algorithm has improper fixed parameter settings due to inaccurate a priori information, the Sage-Husa adaptive filtering and the Transition Probability Matrix (TPM) adaptive update set are introduced to become the adaptive IMM algorithm. To solve the lag problem of multi-model switching, the likelihood function value is set as the judgment flag by using the feature that sub-model likelihood function value can quickly respond to the model change trend, and the judgment window is introduced to correct the TPM matrix elements, which effectively improves the model switching speed. Finally, based on the improved adaptive IMM algorithm, the fusion filtering of four sensor positioning information is carried out to realize the high-precision combined positioning of high-speed trains. Simulation results show that the enhanced algorithm improves the positioning accuracy by 1.6%～14.7% compared with other adaptive IMM algorithms, and it can effectively reduce the peak positional error by increasing the switching speed between models, and it also has a better anti-noise performance.
2024, 46(3): 826-834.
doi: 10.11999/JEIT230248
Abstract:
Orthogonal Time Sequency Multiplexing (OTSM) is a low-complexity modulation method suitable for high-speed mobile scenarios. However, a single waveform design method is difficult to meet diverse application requirements and performance demands. Therefore, on basis of Weighted FRactional Fourier Transform (WFRFT), a Weighted FRactional Walsh-Hadamard Transform (WFRWHT) is proposed and an integrated WFRFT-WFRWHT-OTSM waveform framework based on multidimensional extensions is put forward. Through the flexible configuration of two-dimensional parameters, this framework can be degraded to different waveforms such as OTSM, orthogonal time-frequency-space, hybrid carrier, orthogonal frequency division multiplexing and single carrier. In addition, Bit Error Rate (BER) and Peak-to-Average Power Ratio (PAPR) performances of the integrated WFRFT-WFRWHT-OTSM framework over delay-Doppler channels are studied with Gauss-Seidel (GS) iteration equalization. Simulation results show that the proposed framework achieves better BER and PAPR performances through changing the order of WFRFT and WFRWHT over different delay-Doppler channels.
Orthogonal Time Sequency Multiplexing (OTSM) is a low-complexity modulation method suitable for high-speed mobile scenarios. However, a single waveform design method is difficult to meet diverse application requirements and performance demands. Therefore, on basis of Weighted FRactional Fourier Transform (WFRFT), a Weighted FRactional Walsh-Hadamard Transform (WFRWHT) is proposed and an integrated WFRFT-WFRWHT-OTSM waveform framework based on multidimensional extensions is put forward. Through the flexible configuration of two-dimensional parameters, this framework can be degraded to different waveforms such as OTSM, orthogonal time-frequency-space, hybrid carrier, orthogonal frequency division multiplexing and single carrier. In addition, Bit Error Rate (BER) and Peak-to-Average Power Ratio (PAPR) performances of the integrated WFRFT-WFRWHT-OTSM framework over delay-Doppler channels are studied with Gauss-Seidel (GS) iteration equalization. Simulation results show that the proposed framework achieves better BER and PAPR performances through changing the order of WFRFT and WFRWHT over different delay-Doppler channels.
2024, 46(3): 835-847.
doi: 10.11999/JEIT230261
Abstract:
Due to the extreme shortage of licensed spectrum in cellular systems currently, unlicensed spectrum is thus recommended for cellular systems. The Unmanned Aerial Vehicle(UAV) flight trajectory and power control have a significant impact on spectrum utilization efficiency. However, Three-Dimensional(3D) flight trajectory and power optimization methods based on full spectrum sharing have rarely been unstudied. Therefore, a full spectrum sharing method is first proposed in this paper, where the UAV can use the unlicensed spectrum by controlling the transmission power of uplink cellular users and Device to Device (D2D) users without affecting the data transmission of WiFi devices; at the same time, the UAV can use the licensed spectrum without affecting other downlink cellular users. And then, based on the proposed full spectrum sharing method, a joint optimization problem of 3D flight trajectory and transmission power is constructed under the energy constraint of the UAV battery. In order to solve the proposed complex non-convex optimization problem with multi-variable coupling, the successive convex approximation technique and block coordinate descent method are used to transform the original problem into two convex optimization subproblems i.e. 3D trajectory optimization and power control and solve them iteratively. A large of simulation results show that the proposed spectrum sharing method based on 3D trajectory and power optimization significantly improves the spectrum utilization efficiency.
Due to the extreme shortage of licensed spectrum in cellular systems currently, unlicensed spectrum is thus recommended for cellular systems. The Unmanned Aerial Vehicle(UAV) flight trajectory and power control have a significant impact on spectrum utilization efficiency. However, Three-Dimensional(3D) flight trajectory and power optimization methods based on full spectrum sharing have rarely been unstudied. Therefore, a full spectrum sharing method is first proposed in this paper, where the UAV can use the unlicensed spectrum by controlling the transmission power of uplink cellular users and Device to Device (D2D) users without affecting the data transmission of WiFi devices; at the same time, the UAV can use the licensed spectrum without affecting other downlink cellular users. And then, based on the proposed full spectrum sharing method, a joint optimization problem of 3D flight trajectory and transmission power is constructed under the energy constraint of the UAV battery. In order to solve the proposed complex non-convex optimization problem with multi-variable coupling, the successive convex approximation technique and block coordinate descent method are used to transform the original problem into two convex optimization subproblems i.e. 3D trajectory optimization and power control and solve them iteratively. A large of simulation results show that the proposed spectrum sharing method based on 3D trajectory and power optimization significantly improves the spectrum utilization efficiency.
2024, 46(3): 848-857.
doi: 10.11999/JEIT230206
Abstract:
Due to its high flexibility, good safety and all-weather work, ultraviolet light communication is considered to be a potential communication solution for the emergency communication Unmanned Aerial Vehicle (UAV) formation. Based on the Low Energy Adaptive Clustering Hierarchy (LEACH) algorithm, and combined with JAYA intelligent optimization algorithm, a novel routing optimization algorithm Rcomp JAYA LEACH (RJLEACH) is proposed to improve the effective operation time of the ultraviolet light communication UAV formation. The algorithm is applied to optimize the formation routing of ultraviolet light communication UAVs with different structures, and the results obtained by other algorithms are compared and analyzed. The results show that RJLEACH algorithm reduces the residual energy variance between UAV nodes in the cluster head election stage, and the search for the optimal route reduces the energy consumption of inter-cluster communication. Finally, the time of the first node’s death and half nodes’ death in the network are prolonged by 31.8% and 13.8%, respectively compared with the classic LEACH algorithm, and the energy utilization rate is significantly improved, which can gain valuable time for tasks such as disaster relief and emergency communication.
Due to its high flexibility, good safety and all-weather work, ultraviolet light communication is considered to be a potential communication solution for the emergency communication Unmanned Aerial Vehicle (UAV) formation. Based on the Low Energy Adaptive Clustering Hierarchy (LEACH) algorithm, and combined with JAYA intelligent optimization algorithm, a novel routing optimization algorithm Rcomp JAYA LEACH (RJLEACH) is proposed to improve the effective operation time of the ultraviolet light communication UAV formation. The algorithm is applied to optimize the formation routing of ultraviolet light communication UAVs with different structures, and the results obtained by other algorithms are compared and analyzed. The results show that RJLEACH algorithm reduces the residual energy variance between UAV nodes in the cluster head election stage, and the search for the optimal route reduces the energy consumption of inter-cluster communication. Finally, the time of the first node’s death and half nodes’ death in the network are prolonged by 31.8% and 13.8%, respectively compared with the classic LEACH algorithm, and the energy utilization rate is significantly improved, which can gain valuable time for tasks such as disaster relief and emergency communication.
2024, 46(3): 858-866.
doi: 10.11999/JEIT230329
Abstract:
The joint optimization of user pairing, beamforming and phase shifting in a Non-Orthogonal Multiple Access(NOMA) network assisted by Intelligent Reflecting Surface(IRS) is studied in this paper. In the system, each pair is assigned with one beam and the intra-group successive interference cancellation in the signal detection is adopted. A user pairing strategy independent of transmit beamforming and IRS phase shift is proposed, so the user pairing is separated from the joint optimization and the difficulty and complexity in the solving of the optimization is significantly reduced. Then, the transmit beamforming, power allocation and IRS phase shift are jointly optimized to minimize the total transmission power of the base station. The original optimization problem is non-convex. By using the methods such as slack variables, successive convex approximation, semidefinite relaxation and alternative optimization, the original problem is transformed into a convex problem and solved. Simulation results show that the performance of the proposed scheme is better than that of the scheme that one user is assigned with one beam when the number of the transmit antennas is less than that of the users, and is very close to the scheme that one user is assigned with one beam when the number of antennas is more than that of the users, while the computational complexity of the proposed scheme is lower. Compared with the schemes that employ different user pairing strategy, random IRS phase shift, maximum ratio transmission beamforming, or that without IRS, the proposed scheme is always superior in performance.
The joint optimization of user pairing, beamforming and phase shifting in a Non-Orthogonal Multiple Access(NOMA) network assisted by Intelligent Reflecting Surface(IRS) is studied in this paper. In the system, each pair is assigned with one beam and the intra-group successive interference cancellation in the signal detection is adopted. A user pairing strategy independent of transmit beamforming and IRS phase shift is proposed, so the user pairing is separated from the joint optimization and the difficulty and complexity in the solving of the optimization is significantly reduced. Then, the transmit beamforming, power allocation and IRS phase shift are jointly optimized to minimize the total transmission power of the base station. The original optimization problem is non-convex. By using the methods such as slack variables, successive convex approximation, semidefinite relaxation and alternative optimization, the original problem is transformed into a convex problem and solved. Simulation results show that the performance of the proposed scheme is better than that of the scheme that one user is assigned with one beam when the number of the transmit antennas is less than that of the users, and is very close to the scheme that one user is assigned with one beam when the number of antennas is more than that of the users, while the computational complexity of the proposed scheme is lower. Compared with the schemes that employ different user pairing strategy, random IRS phase shift, maximum ratio transmission beamforming, or that without IRS, the proposed scheme is always superior in performance.
2024, 46(3): 867-874.
doi: 10.11999/JEIT230288
Abstract:
A dynamic Residual Sliding Window Decoding (RSWD) algorithm is proposed in this paper to reduce the high bit error rate caused by error propagation in the Sliding Window Decoding (SWD) algorithm for Spatially-Coupled Low-Density Parity-Check (SC-LDPC) codes. By calculating the residuals of edge information before and after updates, RSWD algorithm dynamically updates the edge of the lowest reliability (or equivalently maximum residual value) with priority, resulting in reduction in the frequency of ineffective edge updating and improvement in the convergence speed of sliding window decoding. The simulation results show that compared to the traditional SWD algorithm, the proposed RSWD algorithm successfully decreases the number of error bit at various positions in the window and shows significant effect on suppressing error propagation; in the high Signal-to-Noise Ratio (SNR) region or in the case of small number of decoding iterations, RSWD presents superior bit error rate performance to SWD; Applying dynamic residuals to both Message Reuse(MR) and Window Extension(WE) window decoding algorithms can also obtain similar observations and improve the decoding performance.
A dynamic Residual Sliding Window Decoding (RSWD) algorithm is proposed in this paper to reduce the high bit error rate caused by error propagation in the Sliding Window Decoding (SWD) algorithm for Spatially-Coupled Low-Density Parity-Check (SC-LDPC) codes. By calculating the residuals of edge information before and after updates, RSWD algorithm dynamically updates the edge of the lowest reliability (or equivalently maximum residual value) with priority, resulting in reduction in the frequency of ineffective edge updating and improvement in the convergence speed of sliding window decoding. The simulation results show that compared to the traditional SWD algorithm, the proposed RSWD algorithm successfully decreases the number of error bit at various positions in the window and shows significant effect on suppressing error propagation; in the high Signal-to-Noise Ratio (SNR) region or in the case of small number of decoding iterations, RSWD presents superior bit error rate performance to SWD; Applying dynamic residuals to both Message Reuse(MR) and Window Extension(WE) window decoding algorithms can also obtain similar observations and improve the decoding performance.
2024, 46(3): 875-885.
doi: 10.11999/JEIT230343
Abstract:
To address issues such as reduced communication quality and poor security caused by malicious jamming attacks, eavesdropping, and imperfect channel state information, a robust resource allocation algorithm for Intelligent Reflecting Surface(IRS)-assisted anti-jamming secure communication system is proposed in this paper. Firstly, based on the minimum security rate constraint, maximum transmit power constraint, and IRS phase shift constraint of legitimate users, a robust resource allocation problem is constructed by jointly optimizing the beamforming vectors of the base station, the covariance matrix of artificial noise, and the phase shift matrix of IRS, in the presence of imperfect channel state information of illegal nodes and unknown beamforming vectors of jammers. Secondly, to solve the problem, the original optimisation problem is transformed into an tractable convex optimisation problem using alternating optimisation, the Cauchy-Schwarz inequality, successive convex approximations and Taylor series expansions. The simulation results demonstrate that the proposed algorithm can effectively enhance system security, reduce power consumption, and improve anti-jamming resilience, compared to traditional algorithms. Furthermore, within a certain range of channel errors, it can decrease the probability of confidential interruption by approximately 35%, indicating its strong robustness.
To address issues such as reduced communication quality and poor security caused by malicious jamming attacks, eavesdropping, and imperfect channel state information, a robust resource allocation algorithm for Intelligent Reflecting Surface(IRS)-assisted anti-jamming secure communication system is proposed in this paper. Firstly, based on the minimum security rate constraint, maximum transmit power constraint, and IRS phase shift constraint of legitimate users, a robust resource allocation problem is constructed by jointly optimizing the beamforming vectors of the base station, the covariance matrix of artificial noise, and the phase shift matrix of IRS, in the presence of imperfect channel state information of illegal nodes and unknown beamforming vectors of jammers. Secondly, to solve the problem, the original optimisation problem is transformed into an tractable convex optimisation problem using alternating optimisation, the Cauchy-Schwarz inequality, successive convex approximations and Taylor series expansions. The simulation results demonstrate that the proposed algorithm can effectively enhance system security, reduce power consumption, and improve anti-jamming resilience, compared to traditional algorithms. Furthermore, within a certain range of channel errors, it can decrease the probability of confidential interruption by approximately 35%, indicating its strong robustness.
2024, 46(3): 886-894.
doi: 10.11999/JEIT230239
Abstract:
The base station of a massive Multiple-Input Multiple-Output (MIMO) system is equipped with hundreds of antennas, enhancing the spectral efficiency of the system and increasing the system costs. To address this problem, our research group proposed a Convergence-Guaranteed Multi-Carrier one-bit precoding (CG-MC1bit) iterative algorithm suitable for Orthogonal Frequency-Division Multiplexing (OFDM) downlink massive MIMO systems, which can ensure superior system performance. However, the corresponding computational complexity is high, hindering the practical application of the algorithm in real-time systems. To address this issue, we propose a model-driven, unfolding neural network, which is based on the CG-MC1bit iterative algorithm and introduces trainable parameters to replace high-complexity operations in forward propagation. In particular, we unfold the iterative algorithm into a neural network and introduce trainable parameters to replace high-complexity operations in forward propagation. Simulation results reveal that this method can automatically update parameters. In addition, compared with the traditional precoding algorithms, the proposed method has a higher convergence speed and lower computational complexity.
The base station of a massive Multiple-Input Multiple-Output (MIMO) system is equipped with hundreds of antennas, enhancing the spectral efficiency of the system and increasing the system costs. To address this problem, our research group proposed a Convergence-Guaranteed Multi-Carrier one-bit precoding (CG-MC1bit) iterative algorithm suitable for Orthogonal Frequency-Division Multiplexing (OFDM) downlink massive MIMO systems, which can ensure superior system performance. However, the corresponding computational complexity is high, hindering the practical application of the algorithm in real-time systems. To address this issue, we propose a model-driven, unfolding neural network, which is based on the CG-MC1bit iterative algorithm and introduces trainable parameters to replace high-complexity operations in forward propagation. In particular, we unfold the iterative algorithm into a neural network and introduce trainable parameters to replace high-complexity operations in forward propagation. Simulation results reveal that this method can automatically update parameters. In addition, compared with the traditional precoding algorithms, the proposed method has a higher convergence speed and lower computational complexity.
2024, 46(3): 895-905.
doi: 10.11999/JEIT230145
Abstract:
Jamming pattern recognition is an indispensable part of modern military communication countermeasure. With the emergence of various new malicious jamming patterns in complex electromagnetic environment, the judgment of unknown jamming has become more and more important. Therefore, the jamming pattern recognition algorithm is required to maintain the high-precision recognition of the known jamming, and can also complete the judgment of the unknown jamming to eliminate the influence of the unknown malicious jamming. Based on this, the jamming pattern recognition problem in the presence of unknown jamming as an open set recognition problem is modeled in this paper, and a jamming pattern open set recognition method based on hyperspherical triple coding is proposed. The proposed method uses hyperspherical triples to reduce the dimension of the input time-frequency image to improve the recognition accuracy, and then uses the meta-recognition classifier to accurately complete the open set recognition of the jamming pattern. The simulation results show that the algorithm can efficiently complete the jamming pattern recognition task in open space when the jamming-to-signal ratio is greater than –2 dB.
Jamming pattern recognition is an indispensable part of modern military communication countermeasure. With the emergence of various new malicious jamming patterns in complex electromagnetic environment, the judgment of unknown jamming has become more and more important. Therefore, the jamming pattern recognition algorithm is required to maintain the high-precision recognition of the known jamming, and can also complete the judgment of the unknown jamming to eliminate the influence of the unknown malicious jamming. Based on this, the jamming pattern recognition problem in the presence of unknown jamming as an open set recognition problem is modeled in this paper, and a jamming pattern open set recognition method based on hyperspherical triple coding is proposed. The proposed method uses hyperspherical triples to reduce the dimension of the input time-frequency image to improve the recognition accuracy, and then uses the meta-recognition classifier to accurately complete the open set recognition of the jamming pattern. The simulation results show that the algorithm can efficiently complete the jamming pattern recognition task in open space when the jamming-to-signal ratio is greater than –2 dB.
2024, 46(3): 906-913.
doi: 10.11999/JEIT230297
Abstract:
To essentially overcome the 3 deficiencies of the mainstream Modulation Wideband Converter (MWC )-based undersampling frequency estimator (i.e., over-consumption of undersampling channels, low accuracy of carrier frequency estimation, high sparsity of the source distribution), this paper proposes the phase-difference corrector based on coprime spectral analysis for the carrier frequency estimation of multi-band communication signals. Specifically, by substituting the multi-path MWC undersampling with the 2-path coprime undersampling, the consumption of undersampling channels is greatly reduced; Further, by developing the mapping relationship between the panoramic spectrum peak indices and the IDFT index pairs of coprime analyzers, the phase difference of the adjacent snapshots’ IDFT outputs corresponding to these index pairs can be analytically extracted, thus achieving much higher estimation accuracy compared to the mainstream MWC method. Meanwhile, by means of incorporating the minimum-sized half-decomposition based all-phase filter design into the prototype filter design, a two-path paralleled coprime spectral analyzer can be constructed, which thoroughly gets rid of the dependency of the high sparsity of the source distribution. Numerical results show that, compared to the mainstream MWC method, the proposed spectral corrector’s estimation error is no more than 1/20 of the former, while only consuming less than half of the sample amount.
To essentially overcome the 3 deficiencies of the mainstream Modulation Wideband Converter (MWC )-based undersampling frequency estimator (i.e., over-consumption of undersampling channels, low accuracy of carrier frequency estimation, high sparsity of the source distribution), this paper proposes the phase-difference corrector based on coprime spectral analysis for the carrier frequency estimation of multi-band communication signals. Specifically, by substituting the multi-path MWC undersampling with the 2-path coprime undersampling, the consumption of undersampling channels is greatly reduced; Further, by developing the mapping relationship between the panoramic spectrum peak indices and the IDFT index pairs of coprime analyzers, the phase difference of the adjacent snapshots’ IDFT outputs corresponding to these index pairs can be analytically extracted, thus achieving much higher estimation accuracy compared to the mainstream MWC method. Meanwhile, by means of incorporating the minimum-sized half-decomposition based all-phase filter design into the prototype filter design, a two-path paralleled coprime spectral analyzer can be constructed, which thoroughly gets rid of the dependency of the high sparsity of the source distribution. Numerical results show that, compared to the mainstream MWC method, the proposed spectral corrector’s estimation error is no more than 1/20 of the former, while only consuming less than half of the sample amount.
2024, 46(3): 914-924.
doi: 10.11999/JEIT230356
Abstract:
Current conventional ambient backscatter communication suffers from double fading, obstacle blockage, and limited network capacity. Reconfigurable Intelligent Surface (RIS) as a key candidate technology of 6G has been concerned due to the improvement of signal transmission quality and the enhancement of the transmission performance of communication systems. Based on the above advantages, RIS and full-duplex techniques are introduced into the ambient backscatter communication system, and the beamforming algorithm is designed in an RIS-enhanced full-duplex ambient backscatter communication network with hardware impairments and discrete phase-shift constraints. Firstly, a beamforming optimization problem is formulated to minimize the total transmit power by considering the minimum harvested energy and the quality-of-service constraint of the backscatter nodes, the maximum transmit power constraint of the power station, and the phase shift constraint of the RIS. Then, the original non-convex problem is transformed into a tractable convex optimization problem by using alternating optimization methods, semi-definite relaxation methods, variable substitution, and semi-definite programming. Finally, simulation results show that the proposed algorithm decreases the average power consumption by 7.8% compared to the conventional beamforming method.
Current conventional ambient backscatter communication suffers from double fading, obstacle blockage, and limited network capacity. Reconfigurable Intelligent Surface (RIS) as a key candidate technology of 6G has been concerned due to the improvement of signal transmission quality and the enhancement of the transmission performance of communication systems. Based on the above advantages, RIS and full-duplex techniques are introduced into the ambient backscatter communication system, and the beamforming algorithm is designed in an RIS-enhanced full-duplex ambient backscatter communication network with hardware impairments and discrete phase-shift constraints. Firstly, a beamforming optimization problem is formulated to minimize the total transmit power by considering the minimum harvested energy and the quality-of-service constraint of the backscatter nodes, the maximum transmit power constraint of the power station, and the phase shift constraint of the RIS. Then, the original non-convex problem is transformed into a tractable convex optimization problem by using alternating optimization methods, semi-definite relaxation methods, variable substitution, and semi-definite programming. Finally, simulation results show that the proposed algorithm decreases the average power consumption by 7.8% compared to the conventional beamforming method.
2024, 46(3): 925-933.
doi: 10.11999/JEIT230153
Abstract:
In the omni-directional underwater acoustic communication network, large propagation delay and high packet collision rate seriously affect the network performance. Compared with the omni-directional reception, the sound pressure and vibration velocity of the acoustic vector sensor can form a unilateral directivity through the linear weighted combination, so as to realize the directional reception of signals in a certain direction, and then effectively improve the network spatial multiplexing rate. The network outage probability in the directional reception mode of single vector sensor is analyzed, and its feasibility is also verified. Then, a Directional Reception Low Collision Probability Media Access Control (DRLCP-MAC) protocol is proposed, which uses a directional reception beam handshake mechanism to establish a stable data transmission link, and constructs multiple pairs of parallel communication links through state transition strategy, so as to reduce virtual carrier monitoring range and improve spatial reuse of the network. Simulation results show that compared with Multiple Access Collision Avoidance for Underwater (MACA-U) protocol and Slotted Floor Acquisition Multiple Access (Slotted-FAMA) protocol, the channel access cost of DRLCP-MAC is reduced by 50% and 60%, the network throughput is increased by about 60% and 400%, and the end-to-end delay is reduced by about 50% and 85%.
In the omni-directional underwater acoustic communication network, large propagation delay and high packet collision rate seriously affect the network performance. Compared with the omni-directional reception, the sound pressure and vibration velocity of the acoustic vector sensor can form a unilateral directivity through the linear weighted combination, so as to realize the directional reception of signals in a certain direction, and then effectively improve the network spatial multiplexing rate. The network outage probability in the directional reception mode of single vector sensor is analyzed, and its feasibility is also verified. Then, a Directional Reception Low Collision Probability Media Access Control (DRLCP-MAC) protocol is proposed, which uses a directional reception beam handshake mechanism to establish a stable data transmission link, and constructs multiple pairs of parallel communication links through state transition strategy, so as to reduce virtual carrier monitoring range and improve spatial reuse of the network. Simulation results show that compared with Multiple Access Collision Avoidance for Underwater (MACA-U) protocol and Slotted Floor Acquisition Multiple Access (Slotted-FAMA) protocol, the channel access cost of DRLCP-MAC is reduced by 50% and 60%, the network throughput is increased by about 60% and 400%, and the end-to-end delay is reduced by about 50% and 85%.
2024, 46(3): 934-943.
doi: 10.11999/JEIT230459
Abstract:
The intelligent transportation system that combines Unmanned Aerial Vehicle (UAV) based traffic monitoring and Mobile Edge Computing (MEC) technologies is considered. In order to ensure the timeliness of data and reduce energy consumption in the system, a UAV computation offloading optimization method considering Age of Information (AoI) is proposed. Firstly, the UAV-assisted MEC system model is established to allow the MEC server to cache commonly used applications and provide UAVs with computation offloading, which supports the UAVs to perform traffic monitoring tasks. By jointly optimizing UAV task offloading decisions, UAV uplink and downlink communication bandwidth allocation, and computing resource allocation of offloaded tasks, the total energy consumption of all UAVs and the MEC server is minimized while satisfying constraints of AoI and resource capacities. Secondly, the system energy consumption minimizing problem is a mixed-integer non-convex optimization problem. Discretization and linearization methods are adopted to quickly obtain an approximately optimal solution to the problem. A discrete point generation algorithm is designed to adjust the approximation error. Finally, simulation results show that even for large non-convex problems, the proposed method can quickly obtain approximately optimal solutions and can satisfy constraints of AoI in different task scenarios, minimizing the system energy consumption as much as possible. The simulation results verify the effectiveness of the proposed method.
The intelligent transportation system that combines Unmanned Aerial Vehicle (UAV) based traffic monitoring and Mobile Edge Computing (MEC) technologies is considered. In order to ensure the timeliness of data and reduce energy consumption in the system, a UAV computation offloading optimization method considering Age of Information (AoI) is proposed. Firstly, the UAV-assisted MEC system model is established to allow the MEC server to cache commonly used applications and provide UAVs with computation offloading, which supports the UAVs to perform traffic monitoring tasks. By jointly optimizing UAV task offloading decisions, UAV uplink and downlink communication bandwidth allocation, and computing resource allocation of offloaded tasks, the total energy consumption of all UAVs and the MEC server is minimized while satisfying constraints of AoI and resource capacities. Secondly, the system energy consumption minimizing problem is a mixed-integer non-convex optimization problem. Discretization and linearization methods are adopted to quickly obtain an approximately optimal solution to the problem. A discrete point generation algorithm is designed to adjust the approximation error. Finally, simulation results show that even for large non-convex problems, the proposed method can quickly obtain approximately optimal solutions and can satisfy constraints of AoI in different task scenarios, minimizing the system energy consumption as much as possible. The simulation results verify the effectiveness of the proposed method.
2024, 46(3): 944-951.
doi: 10.11999/JEIT230088
Abstract:
The length feature plays an important role in the ship target identification. A new algorithm for ship length estimation based on the joint observation of wideband composite bistatic radar is presented, which utilize the mono-/bi-static High-Resolution Range Profiles (HRRPs) and the bistatic spatial geometry relation to estimate the actual length of the ship target with only one-time measurement. Then the estimation errors under different conditions and geometrical constructions are analyzed through Monte Carlo simulations. Finally, the proposed method is validated through typical scene simulation experiments. The results show that when the error of HRRP length is less than 5%, the estimation error of the actual ship length is less than 5%, which provides a new idea for feature extraction and identification of ship targets.
The length feature plays an important role in the ship target identification. A new algorithm for ship length estimation based on the joint observation of wideband composite bistatic radar is presented, which utilize the mono-/bi-static High-Resolution Range Profiles (HRRPs) and the bistatic spatial geometry relation to estimate the actual length of the ship target with only one-time measurement. Then the estimation errors under different conditions and geometrical constructions are analyzed through Monte Carlo simulations. Finally, the proposed method is validated through typical scene simulation experiments. The results show that when the error of HRRP length is less than 5%, the estimation error of the actual ship length is less than 5%, which provides a new idea for feature extraction and identification of ship targets.
2024, 46(3): 952-959.
doi: 10.11999/JEIT230347
Abstract:
An ACM-ML (Amplitude Comparison Monopulse-Maximum Likelihood) algorithm that employs a two-dimensional relaxation iterative search for range and velocity inevitably leads to low computational speeds and large amount calculation. Focusing on the problems mentioned above, a method for efficient Multi-target parameter estimation based on a Space-Time Cascaded MonoPulse algorithm (M-STCMP) is proposed in the paper. The algorithm introduces the monopulse technique to the pulse domain for target velocity computing with temporal monopulse, which results in a one-dimensional search for range and significantly reducing the computational burden of a two-dimensional iterative search within the ACM-ML algorithm. Because the temporal monopulse cannot simultaneously match multiple targets of varying velocities across main beams, the M-STCMP algorithm is improved by estimating the velocity in each Doppler cell with the Doppler information of received signals. To suppress energy leakage between targets, estimations produced in the main beams are cascaded and iterated for each target, that results in greater accuracy. Theoretical analysis and simulation verify the effectiveness of the proposed algorithm.
An ACM-ML (Amplitude Comparison Monopulse-Maximum Likelihood) algorithm that employs a two-dimensional relaxation iterative search for range and velocity inevitably leads to low computational speeds and large amount calculation. Focusing on the problems mentioned above, a method for efficient Multi-target parameter estimation based on a Space-Time Cascaded MonoPulse algorithm (M-STCMP) is proposed in the paper. The algorithm introduces the monopulse technique to the pulse domain for target velocity computing with temporal monopulse, which results in a one-dimensional search for range and significantly reducing the computational burden of a two-dimensional iterative search within the ACM-ML algorithm. Because the temporal monopulse cannot simultaneously match multiple targets of varying velocities across main beams, the M-STCMP algorithm is improved by estimating the velocity in each Doppler cell with the Doppler information of received signals. To suppress energy leakage between targets, estimations produced in the main beams are cascaded and iterated for each target, that results in greater accuracy. Theoretical analysis and simulation verify the effectiveness of the proposed algorithm.
2024, 46(3): 960-966.
doi: 10.11999/JEIT230321
Abstract:
A compact dual-band wideband circularly polarized antenna for Radio Frequency IDentification (RFID) reader is designed. The antenna consists of a curved rectangular patch, an L-shaped patch, and a triangular floor, which is fed through a microstrip line. The two radiation patches control the high and low frequency bands independently, and their axial ratio bandwidth can also be adjusted independently. The triangular floor can change the transverse current and the longitudinal current, so as to change the ratio of the transverse current and the longitudinal current to achieve circular polarization performance. The antenna dimensions are 0.92\begin{document}$ {\lambda _0} $\end{document} \begin{document}$ {\lambda _0} $\end{document} \begin{document}$ {\lambda _0} $\end{document} \begin{document}$ {\lambda _0} $\end{document}
A compact dual-band wideband circularly polarized antenna for Radio Frequency IDentification (RFID) reader is designed. The antenna consists of a curved rectangular patch, an L-shaped patch, and a triangular floor, which is fed through a microstrip line. The two radiation patches control the high and low frequency bands independently, and their axial ratio bandwidth can also be adjusted independently. The triangular floor can change the transverse current and the longitudinal current, so as to change the ratio of the transverse current and the longitudinal current to achieve circular polarization performance. The antenna dimensions are 0.92
2024, 46(3): 967-976.
doi: 10.11999/JEIT230165
Abstract:
Non-contact gesture recognition is a new type of human-computer interaction method with broad application prospects. It can be used in Augmented Reality (AR)/Virtual Reality (VR), smart homes, smart medical, etc., and has recently become a research hotspot. Motivated by the need for more precise micro-motion gesture recognition using mm-wave radar in recent years, a novel micro-motion gesture recognition method based on MIMO millimeter wave radar is proposed in this paper. The MilliMeter Wave CAScaded (MMWCAS) radar cascaded with four AWR1243 radar boards is used to collect gesture echoes. Time-frequency analysis is performed on gesture echoes, and the hand target is detected based on the Range-Doppler (RD) map and 3D point cloud. Through data pre-processing, the Range-Time Map (RTM), Doppler-Time Map (DTM), Azimuth-Time Map (ATM) and Elevation-Time Map (ETM) of the gestures are extracted to more comprehensively characterize the motion of the hand gesture. The mixed Feature-Time Maps (FTM) are formed and adopted for the recognition of 12 types of micro-motion gestures. An innovative gesture recognition network based on DenseNet and Convolutional Block Attention Module (CBAM) is designed, and the mixed FTM is used as the input of the network. Experimental results show that the recognition accuracy reaches 99.03%, achieving high-accuracy gesture recognition. It is discovered that the network focuses on the first half of the gesture movement, which improves the recognition accuracy.
Non-contact gesture recognition is a new type of human-computer interaction method with broad application prospects. It can be used in Augmented Reality (AR)/Virtual Reality (VR), smart homes, smart medical, etc., and has recently become a research hotspot. Motivated by the need for more precise micro-motion gesture recognition using mm-wave radar in recent years, a novel micro-motion gesture recognition method based on MIMO millimeter wave radar is proposed in this paper. The MilliMeter Wave CAScaded (MMWCAS) radar cascaded with four AWR1243 radar boards is used to collect gesture echoes. Time-frequency analysis is performed on gesture echoes, and the hand target is detected based on the Range-Doppler (RD) map and 3D point cloud. Through data pre-processing, the Range-Time Map (RTM), Doppler-Time Map (DTM), Azimuth-Time Map (ATM) and Elevation-Time Map (ETM) of the gestures are extracted to more comprehensively characterize the motion of the hand gesture. The mixed Feature-Time Maps (FTM) are formed and adopted for the recognition of 12 types of micro-motion gestures. An innovative gesture recognition network based on DenseNet and Convolutional Block Attention Module (CBAM) is designed, and the mixed FTM is used as the input of the network. Experimental results show that the recognition accuracy reaches 99.03%, achieving high-accuracy gesture recognition. It is discovered that the network focuses on the first half of the gesture movement, which improves the recognition accuracy.
2024, 46(3): 977-985.
doi: 10.11999/JEIT230162
Abstract:
Considering that Relation Network (RN) ignores the global task correlation information, a Few-Shot Learning(FSL)method based on a Task-Aware Relation Network (TARN) for fully using global task correlation information is proposed in this paper. Method class prototype based on global task relationship is created using the Fuzzy C-Mean (FCM) clustering algorithm, and a Task Correlation Attention mechanism (TCA) is designed to improve the one-vs-one evaluation metric in RN for fusing the global task relationship into features. Compared with RN, in the Mini-ImageNet dataset, the classification accuracy of 5-way 1-shot and 5-way 5-shot settings is increased by 8.15% and 7.0% respectively. While in the Tiered-ImageNet dataset, the classification accuracy of 5-way 1-shot and 5-way 5-shot settings is increased by 7.81 and 6.7% respectively. Compared with the position-awareness relation network, in Mini-ImageNet, the classification accuracy of 5-way 1-shot settings is still increased by 1.24%. Compared with other few-shot image classification methods, TARN also achieves the best performance in these two datasets. The combination of the relation network and task correlation can effectively improve the few-shot image classification accuracy.
Considering that Relation Network (RN) ignores the global task correlation information, a Few-Shot Learning(FSL)method based on a Task-Aware Relation Network (TARN) for fully using global task correlation information is proposed in this paper. Method class prototype based on global task relationship is created using the Fuzzy C-Mean (FCM) clustering algorithm, and a Task Correlation Attention mechanism (TCA) is designed to improve the one-vs-one evaluation metric in RN for fusing the global task relationship into features. Compared with RN, in the Mini-ImageNet dataset, the classification accuracy of 5-way 1-shot and 5-way 5-shot settings is increased by 8.15% and 7.0% respectively. While in the Tiered-ImageNet dataset, the classification accuracy of 5-way 1-shot and 5-way 5-shot settings is increased by 7.81 and 6.7% respectively. Compared with the position-awareness relation network, in Mini-ImageNet, the classification accuracy of 5-way 1-shot settings is still increased by 1.24%. Compared with other few-shot image classification methods, TARN also achieves the best performance in these two datasets. The combination of the relation network and task correlation can effectively improve the few-shot image classification accuracy.
2024, 46(3): 986-994.
doi: 10.11999/JEIT230249
Abstract:
Tongue color is one of the most concerning diagnostic features of tongue diagnosis in Traditional Chinese Medicine (TCM). In practical applications, the performance of the model trained from the tongue data acquired by one device is dramatically degraded when applied to other devices due to the data distribution discrepancy. Therefore, in this paper, a few shot domain adaptation tongue color classification method with two-stage meta-learning is proposed. Firstly, a two-stage meta-learning training strategy is proposed to extract domain invariant features from labeled samples in the source domain, and then, the meta-trained network model is fine-tuned using a few labeled data in the target domain, so that the model can quickly adapt to the new sample characteristics in the target domain, improving the generalization ability of the tongue color classification model and avoid overfitting problem. Next, a progressive pseudo label generation strategy is proposed, which uses the meta-trained model to predict the unlabeled samples in the target domain. The prediction results with high confidence are selected and treated as pseudo labels. So high-quality pseudo labels can be gradually generated. Finally, these high-quality pseudo labels are used to train the model, together with the labeled data in the target domain. The tongue color classification model can be obtained. Considering the noisy pseudo labels, the contrast regularization function is adopted, which can effectively suppress the negative impact of noisy samples in the training process and improve the tongue color classification accuracy in the target domain. The experimental results on two self-established TCM tongue color classification datasets show that the classification accuracy of tongue color in the target domain reaches 91.3% when only 20 labeled samples are given in the target domain, which is only 2.05% lower than that of the supervised classification model in the target domain.
Tongue color is one of the most concerning diagnostic features of tongue diagnosis in Traditional Chinese Medicine (TCM). In practical applications, the performance of the model trained from the tongue data acquired by one device is dramatically degraded when applied to other devices due to the data distribution discrepancy. Therefore, in this paper, a few shot domain adaptation tongue color classification method with two-stage meta-learning is proposed. Firstly, a two-stage meta-learning training strategy is proposed to extract domain invariant features from labeled samples in the source domain, and then, the meta-trained network model is fine-tuned using a few labeled data in the target domain, so that the model can quickly adapt to the new sample characteristics in the target domain, improving the generalization ability of the tongue color classification model and avoid overfitting problem. Next, a progressive pseudo label generation strategy is proposed, which uses the meta-trained model to predict the unlabeled samples in the target domain. The prediction results with high confidence are selected and treated as pseudo labels. So high-quality pseudo labels can be gradually generated. Finally, these high-quality pseudo labels are used to train the model, together with the labeled data in the target domain. The tongue color classification model can be obtained. Considering the noisy pseudo labels, the contrast regularization function is adopted, which can effectively suppress the negative impact of noisy samples in the training process and improve the tongue color classification accuracy in the target domain. The experimental results on two self-established TCM tongue color classification datasets show that the classification accuracy of tongue color in the target domain reaches 91.3% when only 20 labeled samples are given in the target domain, which is only 2.05% lower than that of the supervised classification model in the target domain.
2024, 46(3): 995-1004.
doi: 10.11999/JEIT230208
Abstract:
As the basis of many intelligent visual tasks, Multi-Object Tracking (MOT) is a challenging problem in computer vision. Occlusion is a main factor affecting the tracking accuracy. To solve the occlusion problem, in this paper, the strategy of tracking-by-detection is adopted to obtain complete trajectories of targets based on associating tracklets. Meanwhile, to improve the tracking robustness, the tracklet generation problem is transformed into the facility location problem in operations research area and further a submodular optimization based tracklet generation method is proposed. In this method, two complementary features including Histogram of Oriented Gradient (HOG)and Color Name (CN) are integrated to describe the target appearance, and a weighting coefficient is also designed by motion information to improve the matching accuracy. At length, a submodular maximization algorithm with constraints is developed to achieve the global data association by selecting the targets to form the tracklets. By comparative experiments on the benchmark datasets, the proposed method can solve the occlusion problem effectively with guaranteed performance.
As the basis of many intelligent visual tasks, Multi-Object Tracking (MOT) is a challenging problem in computer vision. Occlusion is a main factor affecting the tracking accuracy. To solve the occlusion problem, in this paper, the strategy of tracking-by-detection is adopted to obtain complete trajectories of targets based on associating tracklets. Meanwhile, to improve the tracking robustness, the tracklet generation problem is transformed into the facility location problem in operations research area and further a submodular optimization based tracklet generation method is proposed. In this method, two complementary features including Histogram of Oriented Gradient (HOG)and Color Name (CN) are integrated to describe the target appearance, and a weighting coefficient is also designed by motion information to improve the matching accuracy. At length, a submodular maximization algorithm with constraints is developed to achieve the global data association by selecting the targets to form the tracklets. By comparative experiments on the benchmark datasets, the proposed method can solve the occlusion problem effectively with guaranteed performance.
2024, 46(3): 1005-1012.
doi: 10.11999/JEIT230260
Abstract:
The current audiovisual speech separation model is essentially the simple splicing of video and audio features without fully considering the interrelationship of each modality, resulting in the underutilization of visual information and unsatisfactory separation effects. The article adequately considers the interconnection between visual features and audio features, adopts a multi-headed attention mechanism, and combines the Convolutional Time-domain audio separation Network (Conv-TasNet) and Dual-Path Recurrent Neural Network (DPRNN), the Multi-Head Attention Time Domain AudioVisual Speech Separation (MHATD-AVSS) model is proposed. The audio encoder and the visual encoder are used to obtain the audio features and the lip features of the video, and the multi-head attention mechanism is used to cross-modality fuse the audio features with the visual features to obtain the audiovisual fusion features, which are passed through the DPRNN separation network to obtain the separated speech of different speakers. The Perceptual Evaluation of Speech Quality (PESQ), Short-Time Objective Intelligibility (STOI), and Signal-to-Noise Ratio (SNR) evaluation metrics are used for experimental testing in the VoxCeleb2 dataset. The research shows that when separating the mixed speech of two, three, or four speakers, the SDR improvement of the method in this paper is above 1.87 dB and up to 2.29 dB compared with the traditional separation network. In summary, this article shows that the method can consider the phase information of the audio signal, better use the correlation between visual information and audio information, extract more accurate audio and video characteristics, and obtain better separation effects.
The current audiovisual speech separation model is essentially the simple splicing of video and audio features without fully considering the interrelationship of each modality, resulting in the underutilization of visual information and unsatisfactory separation effects. The article adequately considers the interconnection between visual features and audio features, adopts a multi-headed attention mechanism, and combines the Convolutional Time-domain audio separation Network (Conv-TasNet) and Dual-Path Recurrent Neural Network (DPRNN), the Multi-Head Attention Time Domain AudioVisual Speech Separation (MHATD-AVSS) model is proposed. The audio encoder and the visual encoder are used to obtain the audio features and the lip features of the video, and the multi-head attention mechanism is used to cross-modality fuse the audio features with the visual features to obtain the audiovisual fusion features, which are passed through the DPRNN separation network to obtain the separated speech of different speakers. The Perceptual Evaluation of Speech Quality (PESQ), Short-Time Objective Intelligibility (STOI), and Signal-to-Noise Ratio (SNR) evaluation metrics are used for experimental testing in the VoxCeleb2 dataset. The research shows that when separating the mixed speech of two, three, or four speakers, the SDR improvement of the method in this paper is above 1.87 dB and up to 2.29 dB compared with the traditional separation network. In summary, this article shows that the method can consider the phase information of the audio signal, better use the correlation between visual information and audio information, extract more accurate audio and video characteristics, and obtain better separation effects.
2024, 46(3): 1013-1021.
doi: 10.11999/JEIT230385
Abstract:
A lesion detection method in ultrasound images based on feature feedback mechanism is proposed to realize real-time accurate localization and detection of ultrasound lesions. The proposed method consists of two parts: feature extraction network based on feature feedback mechanism and adaptive detection head based on divide-and-conquer strategy. The feature feedback network fully learns the global context information and local low-level semantic details of ultrasound images through feedback feature selection and weighted fusion calculation to improve the recognition ability of local lesion features. The adaptive detection head performs divide-and-conquer preprocessing on the multi-level features extracted by the feature feedback network. By combining physiological prior knowledge and feature convolution, adaptive modeling of lesion shape and scale features is performed on features at all levels to enhance the detection effect of the detection head on lesions of different sizes under multi-level features. The proposed method is tested on the thyroid ultrasound image dataset, and 70.3% AP, 99.0% AP50 and 88.4% AP75 are obtained. Experimental results show that the proposed algorithm can achieve more accurate real-time detection and positioning of ultrasound image lesions in comparison with mainstream detection algorithm.
A lesion detection method in ultrasound images based on feature feedback mechanism is proposed to realize real-time accurate localization and detection of ultrasound lesions. The proposed method consists of two parts: feature extraction network based on feature feedback mechanism and adaptive detection head based on divide-and-conquer strategy. The feature feedback network fully learns the global context information and local low-level semantic details of ultrasound images through feedback feature selection and weighted fusion calculation to improve the recognition ability of local lesion features. The adaptive detection head performs divide-and-conquer preprocessing on the multi-level features extracted by the feature feedback network. By combining physiological prior knowledge and feature convolution, adaptive modeling of lesion shape and scale features is performed on features at all levels to enhance the detection effect of the detection head on lesions of different sizes under multi-level features. The proposed method is tested on the thyroid ultrasound image dataset, and 70.3% AP, 99.0% AP50 and 88.4% AP75 are obtained. Experimental results show that the proposed algorithm can achieve more accurate real-time detection and positioning of ultrasound image lesions in comparison with mainstream detection algorithm.
2024, 46(3): 1022-1032.
doi: 10.11999/JEIT230353
Abstract:
In traditional single-channel-based feature extraction approaches, features are captured solely based on dependency, while semantic similarity and dependency types between words are ignored. Although some success has been achieved through the graph convolutional network-based approach for sentiment analysis, aggregating both semantic information and syntactic structure features remains challenging, and the gradual loss of semantic features throughout the training process affects the final sentiment classification effect. To prevent the model from misinterpreting relevant sentiment words due to the absence of prior knowledge, the inclusion of external knowledge is recommended to enrich the text. Presently, how to utilize Graph Neural Networks(GNN) to fuse syntactic and semantic features still deserves further research. A multi-channel enhanced graph convolutional network model is proposed in this paper to address the above issues. First, graph convolution operations on syntactic graphs enhanced with sentiment knowledge and dependency types are performed to obtain two syntax-based representations, which are fused with the semantic representations learned through multi-head attention and graph convolution, so that the multi-channel features can be learned complementarily. It is observed from the experimental results that both the accuracy and macro F1 of our model surpass those of the benchmark model on five publicly available datasets. These findings indicate the importance of dependency types and affective knowledge to enhance syntactic graphs and highlight the effectiveness of combining semantic information with syntactic structure.
In traditional single-channel-based feature extraction approaches, features are captured solely based on dependency, while semantic similarity and dependency types between words are ignored. Although some success has been achieved through the graph convolutional network-based approach for sentiment analysis, aggregating both semantic information and syntactic structure features remains challenging, and the gradual loss of semantic features throughout the training process affects the final sentiment classification effect. To prevent the model from misinterpreting relevant sentiment words due to the absence of prior knowledge, the inclusion of external knowledge is recommended to enrich the text. Presently, how to utilize Graph Neural Networks(GNN) to fuse syntactic and semantic features still deserves further research. A multi-channel enhanced graph convolutional network model is proposed in this paper to address the above issues. First, graph convolution operations on syntactic graphs enhanced with sentiment knowledge and dependency types are performed to obtain two syntax-based representations, which are fused with the semantic representations learned through multi-head attention and graph convolution, so that the multi-channel features can be learned complementarily. It is observed from the experimental results that both the accuracy and macro F1 of our model surpass those of the benchmark model on five publicly available datasets. These findings indicate the importance of dependency types and affective knowledge to enhance syntactic graphs and highlight the effectiveness of combining semantic information with syntactic structure.
2024, 46(3): 1033-1042.
doi: 10.11999/JEIT230179
Abstract:
Traditional deep neural networks usually stack deep features in a way such as skip connection, which is easy to cause information redundancy. To improve the utilization of deep feature information, a Deep Feature Differentiation Network (DFDN) is proposed and applied to single image super-resolution. First, multi-scale deep feature differentiation information is extracted and fused by Mutual-Projected Fusion Block (MPFB) to reduce the contextual information loss. Second, a differential feature attention module is proposed to further learn the differences of deep features while expanding the perception field. Third, the modules are connected in a recursive form to increase the network depth and realize feature reuse. The DIV2K dataset is used as the training dataset, and the pre-trained model is tested with four benchmark datasets, and the results are obtained by comparing the reconstructed images with popular algorithms. Extensive experiments show that the algorithm proposed in this study learns richer texture information than existing algorithms and achieves the best rankings in both subjective visualization and quantitative evaluation metrics, which again proves its robustness and superiority.
Traditional deep neural networks usually stack deep features in a way such as skip connection, which is easy to cause information redundancy. To improve the utilization of deep feature information, a Deep Feature Differentiation Network (DFDN) is proposed and applied to single image super-resolution. First, multi-scale deep feature differentiation information is extracted and fused by Mutual-Projected Fusion Block (MPFB) to reduce the contextual information loss. Second, a differential feature attention module is proposed to further learn the differences of deep features while expanding the perception field. Third, the modules are connected in a recursive form to increase the network depth and realize feature reuse. The DIV2K dataset is used as the training dataset, and the pre-trained model is tested with four benchmark datasets, and the results are obtained by comparing the reconstructed images with popular algorithms. Extensive experiments show that the algorithm proposed in this study learns richer texture information than existing algorithms and achieves the best rankings in both subjective visualization and quantitative evaluation metrics, which again proves its robustness and superiority.
2024, 46(3): 1043-1050.
doi: 10.11999/JEIT230274
Abstract:
To address the problems that blurred decision boundaries and low identifiability of features in the rolling bearing Remaining Useful Life (RUL) prediction under cross working conditions, a domain adaptive method with Maximum Classifier Discrepancy network with Orthogonal Constraints (MCD_OC) is proposed. Firstly, the fast Fourier transform is applied to transform the raw vibration signal into the frequency domain signal and input it to the model. Then, Convolutional Neural Network (CNN) and Gate Recurrent Unit (GRU) are used to extract the depth spatiotemporal features of the bearing signal, the source and target domain feature is aligned using the maximum classifier discrepancy, and the orthogonal constraint is applied to constrain target domain features to increase the identifiability between features of unlabeled target domain feature. Finally, comparative experiments are conducted on the prediction of cross working condition RUL predict based on the bearing life dataset to evaluate the method in this work, and the optimal results are obtained in multiple experiments.
To address the problems that blurred decision boundaries and low identifiability of features in the rolling bearing Remaining Useful Life (RUL) prediction under cross working conditions, a domain adaptive method with Maximum Classifier Discrepancy network with Orthogonal Constraints (MCD_OC) is proposed. Firstly, the fast Fourier transform is applied to transform the raw vibration signal into the frequency domain signal and input it to the model. Then, Convolutional Neural Network (CNN) and Gate Recurrent Unit (GRU) are used to extract the depth spatiotemporal features of the bearing signal, the source and target domain feature is aligned using the maximum classifier discrepancy, and the orthogonal constraint is applied to constrain target domain features to increase the identifiability between features of unlabeled target domain feature. Finally, comparative experiments are conducted on the prediction of cross working condition RUL predict based on the bearing life dataset to evaluate the method in this work, and the optimal results are obtained in multiple experiments.
2024, 46(3): 1051-1059.
doi: 10.11999/JEIT230268
Abstract:
Thanks to the development of deep learning technology, object detection techniques have gained wide attention in various vision tasks. However, obtaining bounding box annotations for objects requires high time and labor costs, which hinders the application of object detection technology in practical scenarios. Therefore, a weakly supervised real-time object detection method based on high resolution class activation mapping algorithm is proposed, using only image class labels to reduce the dependence of network on object instance labels. It subdivides object detection into two subtasks: weakly supervised object localization and real-time object detection. In weakly supervised object localization task, a novel High Resolution Class Activation Mapping(HR-CAM) algorithm based on contrastive layer-wise relevance propagation theory is designed. It can obtain high quality class activation maps and generate pseudo detection annotation box. In real-time detection task, Single Shot multibox Detector(SSD) network as object detector is selected and an Object-Aware Loss function(OA-Loss) based on the class activation maps is designed. It can jointly supervise the training process of the SSD network with generated pseudo detection annotation box, to improve the networks' detection performance for objects. The experimental results show that the method proposed in this paper can achieve accurate and efficient object detection on the CUB200 and TJAB52 datasets, verifying the effectiveness and superiority of this method.
Thanks to the development of deep learning technology, object detection techniques have gained wide attention in various vision tasks. However, obtaining bounding box annotations for objects requires high time and labor costs, which hinders the application of object detection technology in practical scenarios. Therefore, a weakly supervised real-time object detection method based on high resolution class activation mapping algorithm is proposed, using only image class labels to reduce the dependence of network on object instance labels. It subdivides object detection into two subtasks: weakly supervised object localization and real-time object detection. In weakly supervised object localization task, a novel High Resolution Class Activation Mapping(HR-CAM) algorithm based on contrastive layer-wise relevance propagation theory is designed. It can obtain high quality class activation maps and generate pseudo detection annotation box. In real-time detection task, Single Shot multibox Detector(SSD) network as object detector is selected and an Object-Aware Loss function(OA-Loss) based on the class activation maps is designed. It can jointly supervise the training process of the SSD network with generated pseudo detection annotation box, to improve the networks' detection performance for objects. The experimental results show that the method proposed in this paper can achieve accurate and efficient object detection on the CUB200 and TJAB52 datasets, verifying the effectiveness and superiority of this method.
2024, 46(3): 1060-1068.
doi: 10.11999/JEIT230185
Abstract:
At present, methods for classifying blockchain smart contracts using deep learning methods are becoming increasingly popular. However, methods based on deep learning often require a large amount of sample label data for supervised model training to achieve high classification performance. A blockchain smart contract classification method based on a two-level twin neural network in a small sample scenario is proposed to address the problem that currently available smart contract datasets have uneven data categories and insufficient labeled data volumes, which can lead to difficulty in model training and poor classification performance. Firstly, by analyzing the characteristics of smart contract data, a two-level twin neural network model that can capture the characteristics of longer contract data is constructed; Then, based on this model, a training strategy and classification method for smart contracts in small sample scenarios are designed. Finally, experimental results show that the classification performance of the proposed method in this paper is superior to the most advanced smart contract classification methods in small sample scenarios, with a classification accuracy of 94.7% and an F1 value of 94.6%. At the same time, this method requires less tag data, requiring only about 20% data from other methods of the same type.
At present, methods for classifying blockchain smart contracts using deep learning methods are becoming increasingly popular. However, methods based on deep learning often require a large amount of sample label data for supervised model training to achieve high classification performance. A blockchain smart contract classification method based on a two-level twin neural network in a small sample scenario is proposed to address the problem that currently available smart contract datasets have uneven data categories and insufficient labeled data volumes, which can lead to difficulty in model training and poor classification performance. Firstly, by analyzing the characteristics of smart contract data, a two-level twin neural network model that can capture the characteristics of longer contract data is constructed; Then, based on this model, a training strategy and classification method for smart contracts in small sample scenarios are designed. Finally, experimental results show that the classification performance of the proposed method in this paper is superior to the most advanced smart contract classification methods in small sample scenarios, with a classification accuracy of 94.7% and an F1 value of 94.6%. At the same time, this method requires less tag data, requiring only about 20% data from other methods of the same type.
2024, 46(3): 1069-1078.
doi: 10.11999/JEIT230175
Abstract:
To enhance the precision of gait emotion recognition by effectively capturing the dependencies between nodes at multiple scales, long distances, and temporal and spatial positions, a novel method comprising three parts is proposed in this paper. Firstly, a partitioned directed spatio-temporal graph construction method is proposed. It connects all frame nodes in a directed manner based on their regions. Secondly, a multi-scale partition aggregation and fusion method is proposed. This method updates the graph nodes using graph deep learning and fuses similar node features. Lastly, a Multi-scale Partition Directed Adaptive Spatio-Temporal Graph Convolutional Neural network (MPDAST-GCN) is proposed. It constructs a graph in the temporal dimension to obtain the features of distant frame nodes and learns the feature data adaptively on each frame. The MPDAST-GCN classifies input data into four emotion types: happy, sad, angry, and normal. Experimental results on the Emotion-Gait dataset demonstrate that the proposed method outperforms state-of-the-art methods by 6% in terms of accuracy.
To enhance the precision of gait emotion recognition by effectively capturing the dependencies between nodes at multiple scales, long distances, and temporal and spatial positions, a novel method comprising three parts is proposed in this paper. Firstly, a partitioned directed spatio-temporal graph construction method is proposed. It connects all frame nodes in a directed manner based on their regions. Secondly, a multi-scale partition aggregation and fusion method is proposed. This method updates the graph nodes using graph deep learning and fuses similar node features. Lastly, a Multi-scale Partition Directed Adaptive Spatio-Temporal Graph Convolutional Neural network (MPDAST-GCN) is proposed. It constructs a graph in the temporal dimension to obtain the features of distant frame nodes and learns the feature data adaptively on each frame. The MPDAST-GCN classifies input data into four emotion types: happy, sad, angry, and normal. Experimental results on the Emotion-Gait dataset demonstrate that the proposed method outperforms state-of-the-art methods by 6% in terms of accuracy.
2024, 46(3): 1079-1088.
doi: 10.11999/JEIT230188
Abstract:
The Transformer model based on attention mechanism demonstrates superior performance. The complexity of the Transformer model includes both quantity and structural complexity, where the structural complexity leads to a mismatch between irregular models and regular hardware, reducing the efficiency of mapping the model to the hardware. Current accelerator research mainly focuses on addressing the complexity in terms of model quantity, but there is limited research on how to tackle the complexity in model structure. A regularized compressed model is proposd to reduce the structural complexity of the model, improving the matching between the model and the hardware, and increasing the efficiency of mapping the model to the hardware. A hardware-friendly model compression method is introduced, which utilizes a rule-based pruning scheme for weight with offset diagonals and simplifies the hardware quantization inference logic.An efficient and flexible hardware architecture is also present, including a pulsatile operation array with weight fixed at the block level, as well as a quasi-distributed storage architecture. This architecture enables efficient mapping of algorithms to the operation array, while achieving high data storage efficiency and reducing data movement. Experimental results show that the proposed approach achieves a compression rate of 93.75% with minimal performance loss. The accelerator implemented on an FPGA can efficiently handle the compressed Transformer model, resulting in energy efficiency improvements of 12.45 times compared to Central Processing Unit (CPU) and 4.17 times compared to Graphics Processing Unit (GPU).n energy efficiency improvements of 12.45 times compared to Central Processing Unit (CPU) and 4.17 times compared to Graphics Processing Unit (GPU).
The Transformer model based on attention mechanism demonstrates superior performance. The complexity of the Transformer model includes both quantity and structural complexity, where the structural complexity leads to a mismatch between irregular models and regular hardware, reducing the efficiency of mapping the model to the hardware. Current accelerator research mainly focuses on addressing the complexity in terms of model quantity, but there is limited research on how to tackle the complexity in model structure. A regularized compressed model is proposd to reduce the structural complexity of the model, improving the matching between the model and the hardware, and increasing the efficiency of mapping the model to the hardware. A hardware-friendly model compression method is introduced, which utilizes a rule-based pruning scheme for weight with offset diagonals and simplifies the hardware quantization inference logic.An efficient and flexible hardware architecture is also present, including a pulsatile operation array with weight fixed at the block level, as well as a quasi-distributed storage architecture. This architecture enables efficient mapping of algorithms to the operation array, while achieving high data storage efficiency and reducing data movement. Experimental results show that the proposed approach achieves a compression rate of 93.75% with minimal performance loss. The accelerator implemented on an FPGA can efficiently handle the compressed Transformer model, resulting in energy efficiency improvements of 12.45 times compared to Central Processing Unit (CPU) and 4.17 times compared to Graphics Processing Unit (GPU).n energy efficiency improvements of 12.45 times compared to Central Processing Unit (CPU) and 4.17 times compared to Graphics Processing Unit (GPU).
2024, 46(3): 1089-1099.
doi: 10.11999/JEIT230302
Abstract:
Nowadays, Unmanned Aerial Vehicles (UAVs) are widely used in military and civilian fields. While UAVs bring convenience, they also bring huge security risks. The detection and identification technology for UAVs has gradually become a research hotspot. The traditional UAV detection method mainly detects UAVs by obtaining radar echoes, UAV sound signals and photoelectric signals. However, such methods are often susceptible to environmental influences and have certain limitations, and cannot accurately locate and identify UAVs. A UAV identification method based on multi-dimensional signal features is proposed in this paper. Firstly, UAV signals from the received wireless signals through the adaptive triangular threshold method are detected and screened, and at the same time the Channel Status Information (CSI) of the acquired wireless signals is analyzed. Then, the Orthogonal Matching Pursuit (OMP) algorithm is used for parameter estimation to obtain the position information of the UAV to locate the UAV. Finally, the box dimension and Radial Integral Bispectrum (RIB) in UAV signals are extracted to classify and identify UAVs. Through experiments, the method's three-dimensional positioning accuracy for UAVs is less than 1 m, and the classification and recognition accuracy for UAVs can reach up to 100%.
Nowadays, Unmanned Aerial Vehicles (UAVs) are widely used in military and civilian fields. While UAVs bring convenience, they also bring huge security risks. The detection and identification technology for UAVs has gradually become a research hotspot. The traditional UAV detection method mainly detects UAVs by obtaining radar echoes, UAV sound signals and photoelectric signals. However, such methods are often susceptible to environmental influences and have certain limitations, and cannot accurately locate and identify UAVs. A UAV identification method based on multi-dimensional signal features is proposed in this paper. Firstly, UAV signals from the received wireless signals through the adaptive triangular threshold method are detected and screened, and at the same time the Channel Status Information (CSI) of the acquired wireless signals is analyzed. Then, the Orthogonal Matching Pursuit (OMP) algorithm is used for parameter estimation to obtain the position information of the UAV to locate the UAV. Finally, the box dimension and Radial Integral Bispectrum (RIB) in UAV signals are extracted to classify and identify UAVs. Through experiments, the method's three-dimensional positioning accuracy for UAVs is less than 1 m, and the classification and recognition accuracy for UAVs can reach up to 100%.
2024, 46(3): 1100-1108.
doi: 10.11999/JEIT230168
Abstract:
Three-Dimensional Convolution Neural Network (3D CNN) and two-stream Convolution Neural Network (two-stream CNN) are commonly-used for human activities recognition, and each has its own advantages. A human activities recognition model with low complexity and high recognition accuracy is proposed by combining the two architectures. Specifically, a Two-stream NonLocal Spatial Temporal Residual Convolution Neural Network based onchannel Pruning (TPNLST-ResCNN) is proposed in this paper. And Spatial Temporal Residual Convolution Neural Networks (ST-ResCNN) are used both in the temporal stream subnetwork and the spatial stream subnetwork. The final recognition results are acquired by fusing the recognition results of the two subnetworks under a mean fusion algorithm. Furthermore, in order to reduce the complexity of the network, a channel pruning scheme for ST-ResCNN is presented to achieve model compression. In order to enable the compressed network to learn the long-distance spatiotemporal dependencies of human activity changes better and improve the recognition accuracy of the network, a nonlocal block is introduced before the first residual spatial temporal convolution block of the pruned network. The experimental results show that the recognition accuracies of the proposed human activity recognition model are 98.33% and 74.63% on the public dataset UCF101 and HMDB51, respectively. Compared with the existed algorithms, the proposed model in this paper has fewer parameters and higher recognition accuracy.
Three-Dimensional Convolution Neural Network (3D CNN) and two-stream Convolution Neural Network (two-stream CNN) are commonly-used for human activities recognition, and each has its own advantages. A human activities recognition model with low complexity and high recognition accuracy is proposed by combining the two architectures. Specifically, a Two-stream NonLocal Spatial Temporal Residual Convolution Neural Network based onchannel Pruning (TPNLST-ResCNN) is proposed in this paper. And Spatial Temporal Residual Convolution Neural Networks (ST-ResCNN) are used both in the temporal stream subnetwork and the spatial stream subnetwork. The final recognition results are acquired by fusing the recognition results of the two subnetworks under a mean fusion algorithm. Furthermore, in order to reduce the complexity of the network, a channel pruning scheme for ST-ResCNN is presented to achieve model compression. In order to enable the compressed network to learn the long-distance spatiotemporal dependencies of human activity changes better and improve the recognition accuracy of the network, a nonlocal block is introduced before the first residual spatial temporal convolution block of the pruned network. The experimental results show that the recognition accuracies of the proposed human activity recognition model are 98.33% and 74.63% on the public dataset UCF101 and HMDB51, respectively. Compared with the existed algorithms, the proposed model in this paper has fewer parameters and higher recognition accuracy.
2024, 46(3): 1109-1118.
doi: 10.11999/JEIT230193
Abstract:
Current Quantum Secret Sharing(QSS) has the drawbacks of high consumption of resource preparation and the security is not stronger. To overcome the above drawbacks, a verifiable quantum secret sharing scheme based on orthogonal product states is proposed, where multiple participants can dynamically join or leave the secret sharing. In the proposed scheme, the particle pairs of product states are divided into two sequences, the first sequence is transmitted among participants, and the previous participant performs the unitary operator to aggregate the shares on it and then transmits it to the next participant; for the other sequence, the last participant(verifier) performs the Oracle operator on the received particles. Afterward, the verifier uses global measurements on the particle pairs to obtain the quadratic residues of the secrets. Finally, learning from the idea of non-single mapping between ciphertext and plaintext in Rabin cipher, the verifier jointly with Alice verifies the correctness of the measurement results and identifies the secrets from the results. Security analysis shows that the proposed scheme can resist common external and internal attacks, and that the verification process is strongly secure. Since the nonlocal orthogonal product states are transmitted separately in two sequences, the security of the secret reconstruction process is enhanced. Performance analysis shows that the proposed scheme has low quantum resource consumption using orthogonal product state as information carriers, and extends the dimension of orthogonal product basis from low dimension to d dimension, and the number of participants can be dynamically increased or decreased, so it provides better flexibility and generality.
Current Quantum Secret Sharing(QSS) has the drawbacks of high consumption of resource preparation and the security is not stronger. To overcome the above drawbacks, a verifiable quantum secret sharing scheme based on orthogonal product states is proposed, where multiple participants can dynamically join or leave the secret sharing. In the proposed scheme, the particle pairs of product states are divided into two sequences, the first sequence is transmitted among participants, and the previous participant performs the unitary operator to aggregate the shares on it and then transmits it to the next participant; for the other sequence, the last participant(verifier) performs the Oracle operator on the received particles. Afterward, the verifier uses global measurements on the particle pairs to obtain the quadratic residues of the secrets. Finally, learning from the idea of non-single mapping between ciphertext and plaintext in Rabin cipher, the verifier jointly with Alice verifies the correctness of the measurement results and identifies the secrets from the results. Security analysis shows that the proposed scheme can resist common external and internal attacks, and that the verification process is strongly secure. Since the nonlocal orthogonal product states are transmitted separately in two sequences, the security of the secret reconstruction process is enhanced. Performance analysis shows that the proposed scheme has low quantum resource consumption using orthogonal product state as information carriers, and extends the dimension of orthogonal product basis from low dimension to d dimension, and the number of participants can be dynamically increased or decreased, so it provides better flexibility and generality.
2024, 46(3): 1119-1127.
doi: 10.11999/JEIT230137
Abstract:
Federated Learning (FL) performs model training based on local training on clients and continuous model parameters interaction between terminals and server, which effectively solving data leakage and privacy risks in centralized machine learning models. However, since multiple malicious terminals participating in FL can achieve adversarial attacks by inputting small perturbations in the process of local learning, and then lead to incorrect results output by the global model. An effective federated defense strategy – SelectiveFL is proposed in this paper. This strategy first establishes a selective federated defense framework, and then updates the uploaded local model on the server on the basis of extracting attack characteristics through adversarial training at the terminals. At the same time, selective aggregation is carried out according to the attack characteristics, and finally multiple adaptive defense models can be obtained. Finally, the proposed defense method is evaluated on several representative benchmark datasets. The experimental results show that compared with the existing research work, the accuracy of the model can be improved by 2% to 11%.
Federated Learning (FL) performs model training based on local training on clients and continuous model parameters interaction between terminals and server, which effectively solving data leakage and privacy risks in centralized machine learning models. However, since multiple malicious terminals participating in FL can achieve adversarial attacks by inputting small perturbations in the process of local learning, and then lead to incorrect results output by the global model. An effective federated defense strategy – SelectiveFL is proposed in this paper. This strategy first establishes a selective federated defense framework, and then updates the uploaded local model on the server on the basis of extracting attack characteristics through adversarial training at the terminals. At the same time, selective aggregation is carried out according to the attack characteristics, and finally multiple adaptive defense models can be obtained. Finally, the proposed defense method is evaluated on several representative benchmark datasets. The experimental results show that compared with the existing research work, the accuracy of the model can be improved by 2% to 11%.
2024, 46(3): 1128-1137.
doi: 10.11999/JEIT230203
Abstract:
Considering the information security problems involved in the transmission and storage of satellite images, a new satellite image encryption algorithm based on chaos theory and DNA dynamic coding is proposed. Firstly, an improved infinite folding chaotic map is proposed, which broadens the chaotic interval of the original infinite folding chaotic map. Then, combined with the improved Chebyshev chaotic map and SHA-256 hash algorithm, the key stream of the encryption algorithm is generated to improve the plaintext sensitivity of the algorithm. Then, the state value of the chaotic system is used to encode the pixels after Hilbert local scrambling to realize DNA dynamic coding, which solves the weakness of being vulnerable to violent attacks caused by fewer DNA coding rules. Finally, the chaotic sequence is used to complete further chaotic encryption, to completely confuse the original pixel information, increase the randomness and complexity of the encryption algorithm, and obtain the ciphertext image. The experimental results show that the algorithm has a better encryption effect and the ability to deal with various attacks.
Considering the information security problems involved in the transmission and storage of satellite images, a new satellite image encryption algorithm based on chaos theory and DNA dynamic coding is proposed. Firstly, an improved infinite folding chaotic map is proposed, which broadens the chaotic interval of the original infinite folding chaotic map. Then, combined with the improved Chebyshev chaotic map and SHA-256 hash algorithm, the key stream of the encryption algorithm is generated to improve the plaintext sensitivity of the algorithm. Then, the state value of the chaotic system is used to encode the pixels after Hilbert local scrambling to realize DNA dynamic coding, which solves the weakness of being vulnerable to violent attacks caused by fewer DNA coding rules. Finally, the chaotic sequence is used to complete further chaotic encryption, to completely confuse the original pixel information, increase the randomness and complexity of the encryption algorithm, and obtain the ciphertext image. The experimental results show that the algorithm has a better encryption effect and the ability to deal with various attacks.
2024, 46(3): 1138-1146.
doi: 10.11999/JEIT230283
Abstract:
Neurons are the basic unit of the nervous system, and the accuracy of neuron models affects the analysis and understanding of their essential properties. In this paper, a fractional-order photosensitive FitzHugh-Nagumo (FHN) neuron circuit constructed by fractional-order capacitor and inductor is investigated. The dynamics of the fractional-order photosensitive neuron model are analyzed using bifurcation diagrams, phase portraits, and time series diagrams. It was found that the activity of the fractional-order photosensitive neuron increased as the fractional-order decreased. When different system parameters are selected, the neuron system transitions between periodic and chaotic discharge states. The system can induce different discharge modes, such as periodic discharge states, chaotic discharge states, and spike discharge states. In addition, two fractional-order photosensitive neurons were connected using electrical synaptic coupling. Phase synchronization and complete synchronization between the fractional-order photosensitive neuron systems can be achieved by adjusting the coupling strength. Finally, the modulation effect of an external light signal on neuronal excitability was verified by dSPACE.
Neurons are the basic unit of the nervous system, and the accuracy of neuron models affects the analysis and understanding of their essential properties. In this paper, a fractional-order photosensitive FitzHugh-Nagumo (FHN) neuron circuit constructed by fractional-order capacitor and inductor is investigated. The dynamics of the fractional-order photosensitive neuron model are analyzed using bifurcation diagrams, phase portraits, and time series diagrams. It was found that the activity of the fractional-order photosensitive neuron increased as the fractional-order decreased. When different system parameters are selected, the neuron system transitions between periodic and chaotic discharge states. The system can induce different discharge modes, such as periodic discharge states, chaotic discharge states, and spike discharge states. In addition, two fractional-order photosensitive neurons were connected using electrical synaptic coupling. Phase synchronization and complete synchronization between the fractional-order photosensitive neuron systems can be achieved by adjusting the coupling strength. Finally, the modulation effect of an external light signal on neuronal excitability was verified by dSPACE.
