Email alert
2023 Vol. 45, No. 4
Display Method:
2023, 45(4): 1147-1153.
doi: 10.11999/JEIT220170
Abstract:
In a traditional Non-Orthogonal Multiple Access (NOMA) system, more power is usually allocated to edge users to ensure its communication quality. However, the fairness of the system comes at the expense of system capacity. Introducing collaborative communication into the NOMA system, the central user also needs to assume the role of relay in the collaboration phase. This method will inevitably bring a certain burden to the central user. In order to balance system capacity and fairness, a new resource allocation scheme based on cooperative communication and Simultaneous Wireless Information and Power Transfer (SWIPT) is proposed. Energy harvesting equipment is used for energy harvesting, and maximizes the energy efficiency of the system by solving the target problem through Successive Convex Approximation (SCA). Compared with the traditional NOMA and Cooperative NOMA, the energy efficiency of the CNOMA-SWIPT system is greatly improved. When the maximum transmit power of the base station is 30 dBm, CNOMA-SWIPT can achieve a gain of 60.8% compared to the NOMA system and can achieve a gain of about 11.5% higher than that of the CNOMA system, which is more in line with the development concept of green communication.
In a traditional Non-Orthogonal Multiple Access (NOMA) system, more power is usually allocated to edge users to ensure its communication quality. However, the fairness of the system comes at the expense of system capacity. Introducing collaborative communication into the NOMA system, the central user also needs to assume the role of relay in the collaboration phase. This method will inevitably bring a certain burden to the central user. In order to balance system capacity and fairness, a new resource allocation scheme based on cooperative communication and Simultaneous Wireless Information and Power Transfer (SWIPT) is proposed. Energy harvesting equipment is used for energy harvesting, and maximizes the energy efficiency of the system by solving the target problem through Successive Convex Approximation (SCA). Compared with the traditional NOMA and Cooperative NOMA, the energy efficiency of the CNOMA-SWIPT system is greatly improved. When the maximum transmit power of the base station is 30 dBm, CNOMA-SWIPT can achieve a gain of 60.8% compared to the NOMA system and can achieve a gain of about 11.5% higher than that of the CNOMA system, which is more in line with the development concept of green communication.
2023, 45(4): 1154-1162.
doi: 10.11999/JEIT220279
Abstract:
In recent years new services such as road monitoring and assisted driving in smart highways have been proposed, but the explosive growth of data traffic has also emerged, which has brought a great test to the carrying capacity of the network. With the maturity of 5G and mobile edge computing technology, massive tasks do not have to be processed centrally in the cloud, and edge-side co-processing becomes a better choice. In order to provide efficient and reliable services for users in the vehicle high-speed mobile scenario, a Collaboration of Edge Tasks based on Location Prediction (CETLP) is proposed in this paper. First, a delay and load balancing-oriented edge task collaboration model is established by combining the vehicle movement characteristics in the smart highway scenario. Then, a deep reinforcement learning-based edge task collaboration algorithm is proposed to solve the collaboration strategy for a large number of tasks with the objectives of task delay minimization and network load balancing. Simulation results show that the proposed mechanism can reduce the service delay while ensuring the network load balancing.
In recent years new services such as road monitoring and assisted driving in smart highways have been proposed, but the explosive growth of data traffic has also emerged, which has brought a great test to the carrying capacity of the network. With the maturity of 5G and mobile edge computing technology, massive tasks do not have to be processed centrally in the cloud, and edge-side co-processing becomes a better choice. In order to provide efficient and reliable services for users in the vehicle high-speed mobile scenario, a Collaboration of Edge Tasks based on Location Prediction (CETLP) is proposed in this paper. First, a delay and load balancing-oriented edge task collaboration model is established by combining the vehicle movement characteristics in the smart highway scenario. Then, a deep reinforcement learning-based edge task collaboration algorithm is proposed to solve the collaboration strategy for a large number of tasks with the objectives of task delay minimization and network load balancing. Simulation results show that the proposed mechanism can reduce the service delay while ensuring the network load balancing.
2023, 45(4): 1163-1172.
doi: 10.11999/JEIT220140
Abstract:
Intelligent vehicle localization based on 3D Light Detection And Ranging (LiDAR) is still a challenging task in map storage and the efficiency and accuracy of map matching. A lightweight node-level polarized LiDAR map is constructed by a series of nodes with a 2D polarized LiDAR image, a polarized LiDAR fingerprint, and sensor pose, while the polarized LiDAR image encodes a 3D cloud using a multi-channel image format, and the fingerprint is extracted and trained using Siamese network. An intelligent vehicle localization method is also proposed by matching with the polarized LiDAR map. Firstly, Siamese network is used to model the similarity of the query and map fingerprints for fast and coarse map matching. Then a Second-Order Hidden Markov Model (HMM2)-based map sequence matching method is used to find the nearest map node. Finally, the vehicle is readily localized using 3D registration. The proposed method is tested using the actual field data and the public KITTI database. The results indicate that the proposed method can achieve map matching accuracy up to 96% and 30cm localization accuracy with robustness in different types of LiDAR sensors and different environments.
Intelligent vehicle localization based on 3D Light Detection And Ranging (LiDAR) is still a challenging task in map storage and the efficiency and accuracy of map matching. A lightweight node-level polarized LiDAR map is constructed by a series of nodes with a 2D polarized LiDAR image, a polarized LiDAR fingerprint, and sensor pose, while the polarized LiDAR image encodes a 3D cloud using a multi-channel image format, and the fingerprint is extracted and trained using Siamese network. An intelligent vehicle localization method is also proposed by matching with the polarized LiDAR map. Firstly, Siamese network is used to model the similarity of the query and map fingerprints for fast and coarse map matching. Then a Second-Order Hidden Markov Model (HMM2)-based map sequence matching method is used to find the nearest map node. Finally, the vehicle is readily localized using 3D registration. The proposed method is tested using the actual field data and the public KITTI database. The results indicate that the proposed method can achieve map matching accuracy up to 96% and 30cm localization accuracy with robustness in different types of LiDAR sensors and different environments.
2023, 45(4): 1173-1182.
doi: 10.11999/JEIT220231
Abstract:
As a short-range communication technology, Device-to-Device (D2D) communication can greatly reduce the load pressure on cellular base stations and improve spectrum utilization. However, the direct deployment of D2D to licensed or unlicensed bands will inevitably lead to serious interference with existing users. At present, the resource allocation of D2D communication jointly deployed in licensed and unlicensed bands is usually modeled as a mixed-integer nonlinear constraint combinatorial optimization problem, which is difficult to solve by traditional optimization methods. To address this challenging problem, a multi-agent deep reinforcement learning based joint resource allocation D2D communication method is proposed. In this algorithm, each D2D transmitter in the cellular network acts as an agent, which can intelligently select access to the unlicensed channel or the optimal licensed channel and it transmits power through the deep reinforcement learning method. Through the feedback of D2D pairs that compete for the unlicensed channels based on the Listen Before Talk (LBT) mechanism, WiFi network throughput information can be obtained by cellular base station in a non-cooperative manner, so that the algorithm can be executed in a heterogeneous environment and QoS of WiFi users is guaranteed. Compared with Multi Agent Deep Q Network (MADQN), Multi Agent Q Learning (MAQL) and Random Baseline algorithms, the proposed algorithm can achieve the maximum throughput while the QoS is guaranteed for both WiFi users and cellular users.
As a short-range communication technology, Device-to-Device (D2D) communication can greatly reduce the load pressure on cellular base stations and improve spectrum utilization. However, the direct deployment of D2D to licensed or unlicensed bands will inevitably lead to serious interference with existing users. At present, the resource allocation of D2D communication jointly deployed in licensed and unlicensed bands is usually modeled as a mixed-integer nonlinear constraint combinatorial optimization problem, which is difficult to solve by traditional optimization methods. To address this challenging problem, a multi-agent deep reinforcement learning based joint resource allocation D2D communication method is proposed. In this algorithm, each D2D transmitter in the cellular network acts as an agent, which can intelligently select access to the unlicensed channel or the optimal licensed channel and it transmits power through the deep reinforcement learning method. Through the feedback of D2D pairs that compete for the unlicensed channels based on the Listen Before Talk (LBT) mechanism, WiFi network throughput information can be obtained by cellular base station in a non-cooperative manner, so that the algorithm can be executed in a heterogeneous environment and QoS of WiFi users is guaranteed. Compared with Multi Agent Deep Q Network (MADQN), Multi Agent Q Learning (MAQL) and Random Baseline algorithms, the proposed algorithm can achieve the maximum throughput while the QoS is guaranteed for both WiFi users and cellular users.
2023, 45(4): 1183-1191.
doi: 10.11999/JEIT220121
Abstract:
To improve further the concealment performance of satellite signals, combined with the constellation confusion characteristics of Weighted FRactional Fourier Transform (WFRFT) and the anti-interception characteristics of chaotic trajectory, a dual polarization satellite joint modulation scheme based on physical layer security is proposed. Referring to the idea of phase scrambling code and the concept of joint design, the diversity of satellite signal processing is increased by expanding the weighted term parameters of 4-WFRFT. On this basis, logistic mapping encryption is embedded to enhance the difficulty of information cracking. The chaotic secret transmission model of secure dual-polarized satellite communication system is established, a new concept of integrated multi-dimensional concealment based on MP-WFRFT is proposed, and the optimal design and fission mechanism of dual-polarized satellite signal constellation are explored. Simulation results verify the effectiveness of our proposed scheme.
To improve further the concealment performance of satellite signals, combined with the constellation confusion characteristics of Weighted FRactional Fourier Transform (WFRFT) and the anti-interception characteristics of chaotic trajectory, a dual polarization satellite joint modulation scheme based on physical layer security is proposed. Referring to the idea of phase scrambling code and the concept of joint design, the diversity of satellite signal processing is increased by expanding the weighted term parameters of 4-WFRFT. On this basis, logistic mapping encryption is embedded to enhance the difficulty of information cracking. The chaotic secret transmission model of secure dual-polarized satellite communication system is established, a new concept of integrated multi-dimensional concealment based on MP-WFRFT is proposed, and the optimal design and fission mechanism of dual-polarized satellite signal constellation are explored. Simulation results verify the effectiveness of our proposed scheme.
2023, 45(4): 1192-1200.
doi: 10.11999/JEIT220266
Abstract:
In order to improve the safety performance of satellite communication signal, a multi-layer multi-parameter Multi-term Weighted-type FRactional Fourier Transform (MWFRFT) composite modulation communication signal design method is proposed in this paper. Considering the scanning threat of traditional term MWFRFT single-layer structure, MWFRFT is extended to multi-layer structure with different weighting coefficients, which reduces the scanned probability of the system. At the same time, the simulation of communication signal modulation characteristics under multi-layer structure is solved by optimizing the set of control parameters of multi-layer Multi-Parameter MWFRFT (MPMWFRFT) system. Considering the target parasitic signal and narrowband signal interference in complex electromagnetic environment scene, the multi-layer multi-parameter composite modulation system Three Layer Multinomial Weighted FRactional Fourier Transform and Direct Sequence Spread Spectrum (TL-MWFRFT-DSSS) are designed by introducing the spread spectrum mechanism. Simulation results show that this method realizes the simulation of modulation characteristics of multi-layer communication signals on the premise of ensuring good communication performance, and improves significantly the anti scanning performance of the system.
In order to improve the safety performance of satellite communication signal, a multi-layer multi-parameter Multi-term Weighted-type FRactional Fourier Transform (MWFRFT) composite modulation communication signal design method is proposed in this paper. Considering the scanning threat of traditional term MWFRFT single-layer structure, MWFRFT is extended to multi-layer structure with different weighting coefficients, which reduces the scanned probability of the system. At the same time, the simulation of communication signal modulation characteristics under multi-layer structure is solved by optimizing the set of control parameters of multi-layer Multi-Parameter MWFRFT (MPMWFRFT) system. Considering the target parasitic signal and narrowband signal interference in complex electromagnetic environment scene, the multi-layer multi-parameter composite modulation system Three Layer Multinomial Weighted FRactional Fourier Transform and Direct Sequence Spread Spectrum (TL-MWFRFT-DSSS) are designed by introducing the spread spectrum mechanism. Simulation results show that this method realizes the simulation of modulation characteristics of multi-layer communication signals on the premise of ensuring good communication performance, and improves significantly the anti scanning performance of the system.
2023, 45(4): 1201-1210.
doi: 10.11999/JEIT220278
Abstract:
Non-Orthogonal Multiple Access (NOMA) is one of the key candidate technologies for 5G networks. Combined with Cognitive Radio (CR) technology, the CR-NOMA system achieves higher spectral efficiency and greater throughput quantity. The Coordinated Direct and Relay Transmission (CDRT) technology is introduced into the CR-NOMA system in this paper, where one Secondary Source (SS) communicates directly with one near-end secondary user, and the SS can only communicate with multiple far-end secondary users with the help of one Relay (R) node. With imperfect self-interference cancellation and Full-Duplex (FD) relaying, closed-form expressions for the Outage Probability (OP) of the NOMA users are derived. In addition, the revenue optimization problem of SS, R, and the users is analyzed, and an iterative algorithm of two-step user power allocation is proposed. Simulation results show, under the condition of a high signal-to-noise ratio (30 dB), compared with the average power allocation scheme and the random power allocation scheme, the proposed algorithm increases the user sum rate by up to 13%, the SS total income by 56%, and the R total income by 26%. Monte Carlo simulations verify that the theoretical analysis well matches the experimental results.
Non-Orthogonal Multiple Access (NOMA) is one of the key candidate technologies for 5G networks. Combined with Cognitive Radio (CR) technology, the CR-NOMA system achieves higher spectral efficiency and greater throughput quantity. The Coordinated Direct and Relay Transmission (CDRT) technology is introduced into the CR-NOMA system in this paper, where one Secondary Source (SS) communicates directly with one near-end secondary user, and the SS can only communicate with multiple far-end secondary users with the help of one Relay (R) node. With imperfect self-interference cancellation and Full-Duplex (FD) relaying, closed-form expressions for the Outage Probability (OP) of the NOMA users are derived. In addition, the revenue optimization problem of SS, R, and the users is analyzed, and an iterative algorithm of two-step user power allocation is proposed. Simulation results show, under the condition of a high signal-to-noise ratio (30 dB), compared with the average power allocation scheme and the random power allocation scheme, the proposed algorithm increases the user sum rate by up to 13%, the SS total income by 56%, and the R total income by 26%. Monte Carlo simulations verify that the theoretical analysis well matches the experimental results.
2023, 45(4): 1211-1218.
doi: 10.11999/JEIT220246
Abstract:
For evaluating the problem of information security transmission in cooperative Non-Orthogonal Multiple Access (NOMA) system, a Physical Layer Secure (PLS) transmission scheme based on Modify-and-Forward (MF) relay is proposed. The scheme uses MF relay to modify the decoded information and then forward it, to avoid disclosing the legitimate information to the eavesdropping node. Firstly, the NOMA-MF system is modeled, and then Secrecy Outage Probability (SOP), Strictly Positive Security Capacity (SPSC) and Intercept Probability (IP) are deduced to measure the confidentiality and security of the system and Outage Probability (OP) to measure its reliability. In addition, the asymptotic performance of the system is derived, and the performance of NOMA-DF system and NOMA-MF system under Decode-and-Forward (DF) protocol is compared. The derivation and the simulation results show that: The confidentiality and security performance of NOMA-MF system has more advantages than that of NOMA-DF system; There is an optimal Signal-to-Noise Ratio (SNR) between OP and IP of NOMA-MF system to achieve the balance of system safety and reliability; There is an optimal power allocation parameter to achieve the lowest SOP and OP.
For evaluating the problem of information security transmission in cooperative Non-Orthogonal Multiple Access (NOMA) system, a Physical Layer Secure (PLS) transmission scheme based on Modify-and-Forward (MF) relay is proposed. The scheme uses MF relay to modify the decoded information and then forward it, to avoid disclosing the legitimate information to the eavesdropping node. Firstly, the NOMA-MF system is modeled, and then Secrecy Outage Probability (SOP), Strictly Positive Security Capacity (SPSC) and Intercept Probability (IP) are deduced to measure the confidentiality and security of the system and Outage Probability (OP) to measure its reliability. In addition, the asymptotic performance of the system is derived, and the performance of NOMA-DF system and NOMA-MF system under Decode-and-Forward (DF) protocol is compared. The derivation and the simulation results show that: The confidentiality and security performance of NOMA-MF system has more advantages than that of NOMA-DF system; There is an optimal Signal-to-Noise Ratio (SNR) between OP and IP of NOMA-MF system to achieve the balance of system safety and reliability; There is an optimal power allocation parameter to achieve the lowest SOP and OP.
2023, 45(4): 1219-1226.
doi: 10.11999/JEIT220075
Abstract:
In order to improve the bit error rate performance of the irregular Quasi-Cyclic Low-Density Parity-Check (QC-LDPC) codes and reduce the complexity of the construction algorithm, an optimization algorithm based on basis matrix arrangement is proposed. Firstly, the optimal degree distribution of irregular QC-LDPC codes satisfying the code rate and column weight requirements is obtained by using the threshold analysis algorithm based on EXtrinsic Information Transfer (EXIT) chart. Then by using the girth and the number of short-cycles as new indicators, a class of codes with the same optimal degree distribution is analyzed. The basic matrix arrangement structure with the optimal degree distribution and the least number of short-cycles is obtained. Finally, according to the obtained base matrix, the corresponding zeroing operation is performed on the regular index matrix to obtain the target irregular QC-LDPC code. Compared with the random construction method, the proposed construction method has lower implementation complexity. At the same time, the code length and code rate can be flexibly changed by changing the parameter values of the algorithm. Simulation results show that, compared with some existing construction methods, the irregular QC-LDPC codes constructed by the proposed method have better bit error rate performance on Additive White Gaussian Noise (AWGN) channels.
In order to improve the bit error rate performance of the irregular Quasi-Cyclic Low-Density Parity-Check (QC-LDPC) codes and reduce the complexity of the construction algorithm, an optimization algorithm based on basis matrix arrangement is proposed. Firstly, the optimal degree distribution of irregular QC-LDPC codes satisfying the code rate and column weight requirements is obtained by using the threshold analysis algorithm based on EXtrinsic Information Transfer (EXIT) chart. Then by using the girth and the number of short-cycles as new indicators, a class of codes with the same optimal degree distribution is analyzed. The basic matrix arrangement structure with the optimal degree distribution and the least number of short-cycles is obtained. Finally, according to the obtained base matrix, the corresponding zeroing operation is performed on the regular index matrix to obtain the target irregular QC-LDPC code. Compared with the random construction method, the proposed construction method has lower implementation complexity. At the same time, the code length and code rate can be flexibly changed by changing the parameter values of the algorithm. Simulation results show that, compared with some existing construction methods, the irregular QC-LDPC codes constructed by the proposed method have better bit error rate performance on Additive White Gaussian Noise (AWGN) channels.
2023, 45(4): 1227-1235.
doi: 10.11999/JEIT220190
Abstract:
Focusing on the problem that the signal characteristics are not related to the hardware composition in the specific emitter identification, two feature extraction methods are used in this paper: high-order spectral analysis and Variational Modal Decomposition (VMD) for research and analysis. The surrounding-line bispectral integration and improved VMD technology are used to extract and analyze the features of the hardware in the semi-physical simulation signal and Advanced Design System (ADS) output signal. Through ADS, the influence of emitter phase noise and nonlinear distortion of power amplifier on signal unintentional modulation characteristics is quantitatively analyzed, the correlation of variables is analyzed, and the significantly related variables are regressed and fitted to obtain their correlation regression function. Using the correlation between hardware and features, the traditional Support Vector Machines (SVM) classifier is improved to construct a correlation weight SVM classifier. Finally, the results show that the classification accuracy of weighted SVM is improved by more than 10% compared with single core SVM under the same signal-to-noise ratio.
Focusing on the problem that the signal characteristics are not related to the hardware composition in the specific emitter identification, two feature extraction methods are used in this paper: high-order spectral analysis and Variational Modal Decomposition (VMD) for research and analysis. The surrounding-line bispectral integration and improved VMD technology are used to extract and analyze the features of the hardware in the semi-physical simulation signal and Advanced Design System (ADS) output signal. Through ADS, the influence of emitter phase noise and nonlinear distortion of power amplifier on signal unintentional modulation characteristics is quantitatively analyzed, the correlation of variables is analyzed, and the significantly related variables are regressed and fitted to obtain their correlation regression function. Using the correlation between hardware and features, the traditional Support Vector Machines (SVM) classifier is improved to construct a correlation weight SVM classifier. Finally, the results show that the classification accuracy of weighted SVM is improved by more than 10% compared with single core SVM under the same signal-to-noise ratio.
2023, 45(4): 1236-1242.
doi: 10.11999/JEIT220261
Abstract:
In this paper, an integrated detection and tracking algorithm assisted by multi-information is proposed for bistatic radar. Combining with multiple prior information such as target position and echo amplitude obtained by the radar in the target tracking stage assist in the design of the detection threshold within the tracking gate in order to improve the detection and tracking performance of the target. Firstly, according to the prior information of the acquired target location, the traditional likelihood ratio detector is modified based on the Bayesian minimum error criterion under the framework of Probabilistic Data Association (PDA). In order to improve further the weak target detection performance, the track termination criterion is introduced to relax the threshold setting rules, and the average false alarm probability and detection probability within the tracking gate are rededuced. Finally, the calculation method of the association probability of the PDA algorithm are calculated in the case of multi-information assistance. A complete algorithm flow is given, and the feasibility and effectiveness of the algorithm are verified through simulation experiments.
In this paper, an integrated detection and tracking algorithm assisted by multi-information is proposed for bistatic radar. Combining with multiple prior information such as target position and echo amplitude obtained by the radar in the target tracking stage assist in the design of the detection threshold within the tracking gate in order to improve the detection and tracking performance of the target. Firstly, according to the prior information of the acquired target location, the traditional likelihood ratio detector is modified based on the Bayesian minimum error criterion under the framework of Probabilistic Data Association (PDA). In order to improve further the weak target detection performance, the track termination criterion is introduced to relax the threshold setting rules, and the average false alarm probability and detection probability within the tracking gate are rededuced. Finally, the calculation method of the association probability of the PDA algorithm are calculated in the case of multi-information assistance. A complete algorithm flow is given, and the feasibility and effectiveness of the algorithm are verified through simulation experiments.
2023, 45(4): 1243-1253.
doi: 10.11999/JEIT220250
Abstract:
With the development of Synthetic Aperture Radar (SAR) technology, Multidimensional Space Joint-observation, which includes polarimetry, frequency, angle, and time spaces, has become an important trend in SAR development, but there are few reports on systems and experiments about it. In this paper, the capabilities of the airborne Multidimensional Space Joint-observation SAR(MSJosSAR)system are briefly described and the technical characteristics of the system are summarized. A SAR consistently imaging algorithm is proposed and the registration accuracy of Multiband image is better than 1 pixel. The preliminary process of three multidimensional space joint-observations, including multi-band polarization angle feature quantity, multi-aspect tomography three-dimensional reconstruction, are analyzed, multitemporal coherence change detection, which verify the ability of multidimensional space joint-observation.
With the development of Synthetic Aperture Radar (SAR) technology, Multidimensional Space Joint-observation, which includes polarimetry, frequency, angle, and time spaces, has become an important trend in SAR development, but there are few reports on systems and experiments about it. In this paper, the capabilities of the airborne Multidimensional Space Joint-observation SAR(MSJosSAR)system are briefly described and the technical characteristics of the system are summarized. A SAR consistently imaging algorithm is proposed and the registration accuracy of Multiband image is better than 1 pixel. The preliminary process of three multidimensional space joint-observations, including multi-band polarization angle feature quantity, multi-aspect tomography three-dimensional reconstruction, are analyzed, multitemporal coherence change detection, which verify the ability of multidimensional space joint-observation.
2023, 45(4): 1254-1264.
doi: 10.11999/JEIT220322
Abstract:
Airborne Synthetic Aperture Radar Altimeter (SARA) is capable of exploiting the high-resolution in along-track, which has been attracted wide concerns. However, the existing re-tracking methods are mostly based on the least square operator. The performance of estimation accuracy and noise suppression of the operator are limited due to neglect of noise factors and accordance over-fitting problem. In this paper, Parameterized Retracking Bayes (PR-Bayes) algorithm is proposed under the framework of Bayesian machine learning. By introducing a prior probability model of the terrain scene, and combining with model-driven machine learning method, the elevation information of re-tracking with reliable estimation of the target can be achieved. The problem about over-fitting can be alleviated effectively. In this algorithm, Brown Model (BM) is used to recover complicated model parameters of SARA echo. Then, Hamilton Monte Carlo (HMC) statistical sampler is designed to estimate the terrain height of the scene with a high accuracy and reliable confidence. The accuracy and validity of this algorithm are verified by point target simulation and semi-physical simulation based on DEM respectively, and the practicability is proved by the airborne raw SARA data.
Airborne Synthetic Aperture Radar Altimeter (SARA) is capable of exploiting the high-resolution in along-track, which has been attracted wide concerns. However, the existing re-tracking methods are mostly based on the least square operator. The performance of estimation accuracy and noise suppression of the operator are limited due to neglect of noise factors and accordance over-fitting problem. In this paper, Parameterized Retracking Bayes (PR-Bayes) algorithm is proposed under the framework of Bayesian machine learning. By introducing a prior probability model of the terrain scene, and combining with model-driven machine learning method, the elevation information of re-tracking with reliable estimation of the target can be achieved. The problem about over-fitting can be alleviated effectively. In this algorithm, Brown Model (BM) is used to recover complicated model parameters of SARA echo. Then, Hamilton Monte Carlo (HMC) statistical sampler is designed to estimate the terrain height of the scene with a high accuracy and reliable confidence. The accuracy and validity of this algorithm are verified by point target simulation and semi-physical simulation based on DEM respectively, and the practicability is proved by the airborne raw SARA data.
2023, 45(4): 1265-1274.
doi: 10.11999/JEIT220193
Abstract:
The multi-layer structure of the ionosphere can support several signal propagation paths between the sky-wave Over-The-Horizon Radar (OTHR) and targets, often giving rise to multipath measurements for a single target. The problem of multipath measurements clustering for OTHR is considered, which needs to solve the problems of multipath measurements recognition and measurements clustering at the same time. OTHR measurements model assumes that a target can generate at most one measurement through an ionospheric propagation path, and multipath clustering constraints need to be considered. In this paper, affinity propagation clustering is extended to multipath constraint model, and a new multipath constraint affinity propagation clustering algorithm is proposed. The algorithm transforms the clustering problem into an inference problem by constructing the probabilistic graphical model of multipath measurements clustering, and uses the max-sum belief propagation to approximate the maximum a posteriori probability of the clustering matrix. The advantages of the algorithm include that it identifies automatically the number of clusters, and the computational complexity scales quadratically in the number of measurements and the number of propagation paths. Simulation results show that, the proposed method can outperform multiple hypothesis multipath clustering algorithm.
The multi-layer structure of the ionosphere can support several signal propagation paths between the sky-wave Over-The-Horizon Radar (OTHR) and targets, often giving rise to multipath measurements for a single target. The problem of multipath measurements clustering for OTHR is considered, which needs to solve the problems of multipath measurements recognition and measurements clustering at the same time. OTHR measurements model assumes that a target can generate at most one measurement through an ionospheric propagation path, and multipath clustering constraints need to be considered. In this paper, affinity propagation clustering is extended to multipath constraint model, and a new multipath constraint affinity propagation clustering algorithm is proposed. The algorithm transforms the clustering problem into an inference problem by constructing the probabilistic graphical model of multipath measurements clustering, and uses the max-sum belief propagation to approximate the maximum a posteriori probability of the clustering matrix. The advantages of the algorithm include that it identifies automatically the number of clusters, and the computational complexity scales quadratically in the number of measurements and the number of propagation paths. Simulation results show that, the proposed method can outperform multiple hypothesis multipath clustering algorithm.
2023, 45(4): 1275-1284.
doi: 10.11999/JEIT220215
Abstract:
To solve the optimization problem of bistatic radar for belt barrier coverage, a two-dimensional deployment optimization method using adjacent deployment lines is proposed. First, the field of interest is approximately represented by a rectangular area, and then the area is divided into multiple identical sub-barrier coverage areas. To give full play to the efficiency of the transmitters, the bistatic radars consisting of the transmitters and the receivers on the same deployment line and the adjacent deployment lines should be considered. Thus, four two-dimensional deployment patterns are introduced and analyzed. Furthermore, an optimization model based on these deployment patterns is proposed. The minimum deployment cost and the coverage area requirements are adopted as the optimization criterion and constraints, respectively. A method based on the greedy algorithm is exploited to solve the optimization model. The deployment locations and the minimum deployment cost can be calculated by the proposed method. Finally, the simulation results and analysis show that, compared with the existing method, the minimum deployment cost and the number of transmitters can be impressively reduced by the proposed method. The effectiveness of the proposed method is proven.
To solve the optimization problem of bistatic radar for belt barrier coverage, a two-dimensional deployment optimization method using adjacent deployment lines is proposed. First, the field of interest is approximately represented by a rectangular area, and then the area is divided into multiple identical sub-barrier coverage areas. To give full play to the efficiency of the transmitters, the bistatic radars consisting of the transmitters and the receivers on the same deployment line and the adjacent deployment lines should be considered. Thus, four two-dimensional deployment patterns are introduced and analyzed. Furthermore, an optimization model based on these deployment patterns is proposed. The minimum deployment cost and the coverage area requirements are adopted as the optimization criterion and constraints, respectively. A method based on the greedy algorithm is exploited to solve the optimization model. The deployment locations and the minimum deployment cost can be calculated by the proposed method. Finally, the simulation results and analysis show that, compared with the existing method, the minimum deployment cost and the number of transmitters can be impressively reduced by the proposed method. The effectiveness of the proposed method is proven.
2023, 45(4): 1285-1292.
doi: 10.11999/JEIT220239
Abstract:
The Interrupted Frequency Modulated Continuous Wave (IFMCW) Synthetic Aperture Radar (SAR) is a novel type of SAR system,which has the advantages of light weight, low cost, and low power consumption. The system subverts the design concept of the traditional Frequency Modulated Continuous Wave (FMCW) SAR system, which use a single antenna to transmit and receive signals.In this system, the transmitter and receiver operate at different time intervals, resulting in periodic gaps in the synthetic aperture. When the received echo data is imaged using traditional imaging algorithms, the artifacts will appear in the focused SAR image. In order to suppress effectively the appearance of artifacts, this paper proposes a new imaging algorithm for subaperture echo data processing, which is called Low-rank Hankel matrix Reconstruction Technique based on Subaperture Projection (LHRTSP). The experimental results show that the proposed method has better suppression effect on artifacts compared with the existing methods, which verifies the effectiveness and superiority of the proposed method.
The Interrupted Frequency Modulated Continuous Wave (IFMCW) Synthetic Aperture Radar (SAR) is a novel type of SAR system,which has the advantages of light weight, low cost, and low power consumption. The system subverts the design concept of the traditional Frequency Modulated Continuous Wave (FMCW) SAR system, which use a single antenna to transmit and receive signals.In this system, the transmitter and receiver operate at different time intervals, resulting in periodic gaps in the synthetic aperture. When the received echo data is imaged using traditional imaging algorithms, the artifacts will appear in the focused SAR image. In order to suppress effectively the appearance of artifacts, this paper proposes a new imaging algorithm for subaperture echo data processing, which is called Low-rank Hankel matrix Reconstruction Technique based on Subaperture Projection (LHRTSP). The experimental results show that the proposed method has better suppression effect on artifacts compared with the existing methods, which verifies the effectiveness and superiority of the proposed method.
2023, 45(4): 1293-1302.
doi: 10.11999/JEIT220127
Abstract:
In light of difficulties in detecting stable Pulse Repetition Interval (PRI) sequence and estimating its PRI in interleaving pulse stream, a fast detecting method based on plane transformation is proposed to solve the above problems. By one time-domain transform, the Periodic Graph of Plane Transformation Point trace (PGPTP) mapping pulse sequence with stable PRI is formed via the presented method performing integer remainder operation on pulse Time Of Arrival (TOA) over plane width. According to the differences in pattern of PGPTP got previously, pulse sequences with stable PRIs interleaved in TOA can be distinguished, and then the stable PRIs are got with point trace longitudinal deployment period and plane width and are therefore used to accurately deinterleave such sequences in dense pulse stream. Simulation results show that this method is valid and possessed of advantages such as high efficiency and practical applicability, etc.
In light of difficulties in detecting stable Pulse Repetition Interval (PRI) sequence and estimating its PRI in interleaving pulse stream, a fast detecting method based on plane transformation is proposed to solve the above problems. By one time-domain transform, the Periodic Graph of Plane Transformation Point trace (PGPTP) mapping pulse sequence with stable PRI is formed via the presented method performing integer remainder operation on pulse Time Of Arrival (TOA) over plane width. According to the differences in pattern of PGPTP got previously, pulse sequences with stable PRIs interleaved in TOA can be distinguished, and then the stable PRIs are got with point trace longitudinal deployment period and plane width and are therefore used to accurately deinterleave such sequences in dense pulse stream. Simulation results show that this method is valid and possessed of advantages such as high efficiency and practical applicability, etc.
2023, 45(4): 1303-1312.
doi: 10.11999/JEIT220213
Abstract:
High-resolution radar systems monitor multiple extended targets with different shapes in a surveillance area. Reliable shapes estimation can effectively improve tracking performance and are crucial to battle-field situation evaluations. In this paper, a Joint Likelihood based Generalized Labeled Multi-Bernoulli (JL-GLMB) filter is proposed to estimate accurately the number of targets, target tracks, and target shapes. Firstly, the extended target is modeled as a star-convex set, and Gaussian components in the GLMB density are updated by the measurement transformation filter to improve the accuracy of state estimation. Then, a joint likelihood function is constructed by log-weighted fusion strategy to measure comprehensively the similarity between extended target and measurement cell. Finally, a fast approximation method for posterior probability density is proposed based on Gibbs sampling, which improves the accuracy and efficiency of the data association. Simulation results show that the proposed algorithm can effectively estimate multiple extended target states of different shapes, and provide stable cardinality estimation in the clutter environment compared to traditional multiple extended target tracking.
High-resolution radar systems monitor multiple extended targets with different shapes in a surveillance area. Reliable shapes estimation can effectively improve tracking performance and are crucial to battle-field situation evaluations. In this paper, a Joint Likelihood based Generalized Labeled Multi-Bernoulli (JL-GLMB) filter is proposed to estimate accurately the number of targets, target tracks, and target shapes. Firstly, the extended target is modeled as a star-convex set, and Gaussian components in the GLMB density are updated by the measurement transformation filter to improve the accuracy of state estimation. Then, a joint likelihood function is constructed by log-weighted fusion strategy to measure comprehensively the similarity between extended target and measurement cell. Finally, a fast approximation method for posterior probability density is proposed based on Gibbs sampling, which improves the accuracy and efficiency of the data association. Simulation results show that the proposed algorithm can effectively estimate multiple extended target states of different shapes, and provide stable cardinality estimation in the clutter environment compared to traditional multiple extended target tracking.
2023, 45(4): 1313-1320.
doi: 10.11999/JEIT220171
Abstract:
This paper focuses on adaptive filtering techniques for parameter identification of Hammerstein systems and output prediction of nonlinear systems. By formulating the underlying filtering problem as a recursive bilinear least-squares optimization with the non-convex feasible region constraint, an algorithmic framework is developed based on recursive least-squares and Alternating Direction Method of Multipliers (ADMM). Under this framework, the solution to nonconvex constraint optimization problem can be obtained by implementing ridge regression and Euclidean projection. Simulation results in the context of system identification, nonlinear predication, and acoustic echo cancellation, reveal that the proposed algorithm has good convergence and stability, and can obtain more accurate identification and prediction results.
This paper focuses on adaptive filtering techniques for parameter identification of Hammerstein systems and output prediction of nonlinear systems. By formulating the underlying filtering problem as a recursive bilinear least-squares optimization with the non-convex feasible region constraint, an algorithmic framework is developed based on recursive least-squares and Alternating Direction Method of Multipliers (ADMM). Under this framework, the solution to nonconvex constraint optimization problem can be obtained by implementing ridge regression and Euclidean projection. Simulation results in the context of system identification, nonlinear predication, and acoustic echo cancellation, reveal that the proposed algorithm has good convergence and stability, and can obtain more accurate identification and prediction results.
2023, 45(4): 1321-1329.
doi: 10.11999/JEIT220883
Abstract:
Ground-Based Synthetic Aperture Radar (GBSAR) is an all-day all-weather, non-contact, high-precision instrument for wide-area deformation monitoring, which has been widely used to monitormining areas, slops, and dams. When monitoring the outside scene with the radar placed in the inner space, the radar echo would be interfered with by strong scattering signals reflected from the inner space. The strong scattering signal at near range would severely affect the image quality. Therefore, this paper proposes a Robust Principal Component Analysis(RPCA) based algorithm to decompose the range-doppler domain signal into low-rank and sparse parts,as, in the range-doppler domain, the near-range coupled signal has low-rank characteristics, whereas the scene signal has sparse characteristics. Unlike the existing Principal Component Analysis(PCA) based algorithm, the proposed RPCA algorithm does not assume a Gaussian-distributed scene signal, which usually could not be satisfied in reality. Additionally, this paper proposes a correlation-based regularization parameter optimization method for RPCA. Thus, low rank and sparse matrices can be better separated. Furthermore, the proposed method is verified with real GBSAR data. The result shows that the proposed RPCA based method can better suppress the coupled signal while retaining the scene signal than the existing PCA-based algorithm.
2023, 45(4): 1330-1337.
doi: 10.11999/JEIT220067
Abstract:
In order to solve the problem of single function of passive metasurface, a polarization multiplexed transmission Huygens metasurface is proposed, which realizes the independent focusing characteristics of x-polarized and y-polarized incident waves. The metasurface unit consists of a pair of asymmetrical electric dipole elements and a dielectric substrate with a thickness of 0.17 λ. Magnetic dipole is formed by surface current flowing in reverse direction, which eliminates the need for magnetic element in physical structure and makes the unit more compact. Dual-polarization independent control and 360° phase coverage are achieved by adjusting the dimensions of unit. Aligning Huygens particles based on holographic theory, a polarization multiplexed Huygens metasurface with independent focusing characteristics at 35 GHz is designed and fabricated. The measured results are basically consistent with the simulations. The proposed Huygens metasurface has no multilayer stacking and metal via, and has the characteristics of simple structure, low profile, and easy processing.
In order to solve the problem of single function of passive metasurface, a polarization multiplexed transmission Huygens metasurface is proposed, which realizes the independent focusing characteristics of x-polarized and y-polarized incident waves. The metasurface unit consists of a pair of asymmetrical electric dipole elements and a dielectric substrate with a thickness of 0.17 λ. Magnetic dipole is formed by surface current flowing in reverse direction, which eliminates the need for magnetic element in physical structure and makes the unit more compact. Dual-polarization independent control and 360° phase coverage are achieved by adjusting the dimensions of unit. Aligning Huygens particles based on holographic theory, a polarization multiplexed Huygens metasurface with independent focusing characteristics at 35 GHz is designed and fabricated. The measured results are basically consistent with the simulations. The proposed Huygens metasurface has no multilayer stacking and metal via, and has the characteristics of simple structure, low profile, and easy processing.
2023, 45(4): 1338-1345.
doi: 10.11999/JEIT220119
Abstract:
Considering the single defect of the transmitter of the existing mechanical ultra-low frequency antenna and the bottleneck of limited communication distance, In order to realize the miniaturization and long-distance transmission of ultra-low frequency electromagnetic signaling system, the theoretical innovation and engineering practice of low-frequency electromagnetic signaling technology of rotating permanent magnet mechanical antenna are carried out. The electromagnetic radiation theory of ultra-low frequency multi mechanical antenna in Multiple Input Single Output (MISO) scenario is explored, and the spatial magnetic field distribution model of multi mechanical antenna array based on three-phase induction motor is established. The simulation results show that the magnetic induction intensity can be increased by 3 dB in the near field by using the mechanical antenna array composed of three-phase induction motor. The ultra-low frequency near-field optimal beamforming technology with multiple mechanical antennas is proposed, The simulation results show that when the initial phases between the antennas are equal, the field strength of the radial received magnetic field component is the largest. High precision synchronization is designed and principle prototype is setted up for testing, The experimental results show that the signal power increased by 6 dBm and transmission distance can reach 50 m by using binary antenna array.
Considering the single defect of the transmitter of the existing mechanical ultra-low frequency antenna and the bottleneck of limited communication distance, In order to realize the miniaturization and long-distance transmission of ultra-low frequency electromagnetic signaling system, the theoretical innovation and engineering practice of low-frequency electromagnetic signaling technology of rotating permanent magnet mechanical antenna are carried out. The electromagnetic radiation theory of ultra-low frequency multi mechanical antenna in Multiple Input Single Output (MISO) scenario is explored, and the spatial magnetic field distribution model of multi mechanical antenna array based on three-phase induction motor is established. The simulation results show that the magnetic induction intensity can be increased by 3 dB in the near field by using the mechanical antenna array composed of three-phase induction motor. The ultra-low frequency near-field optimal beamforming technology with multiple mechanical antennas is proposed, The simulation results show that when the initial phases between the antennas are equal, the field strength of the radial received magnetic field component is the largest. High precision synchronization is designed and principle prototype is setted up for testing, The experimental results show that the signal power increased by 6 dBm and transmission distance can reach 50 m by using binary antenna array.
2023, 45(4): 1346-1355.
doi: 10.11999/JEIT220207
Abstract:
Considering the problem of multi-keyword fuzzy matching and user fairness in one-to-many data ciphertext sharing, a multi-keyword fuzzy search encryption scheme based on blockchain is proposed. An R-HashMap index structure is proposed. The secure index is constructed by using pairs coding function and position sensitive hash function, and the K-nearest neighbor algorithm is used to encrypt the index. The similarity between the query keyword vector and the index node is calculated by Euclidian distance measure, and the multi-keyword fuzzy ciphertext search is realized. In addition to eliminating the pre-defined dictionary and reducing the storage overhead, this scheme also realizes the update of the security index without increasing the search complexity. In addition, the combination of Ethereum blockchain technology and searchable encryption scheme avoids data tampering by malicious servers, and the use of smart contracts as trusted third parties for retrieval work can not only prevent keyword guessing attacks within cloud servers, but also solve the problem of incorrect retrieval results. Through security proof analysis, the proposed scheme not only satisfies the semantic security of adaptive keyword selection, but also can protect user privacy and data security. The experimental comparison between this program and other schemes proves that the program has better efficiency in terms of time and cost while ensuring accuracy.
Considering the problem of multi-keyword fuzzy matching and user fairness in one-to-many data ciphertext sharing, a multi-keyword fuzzy search encryption scheme based on blockchain is proposed. An R-HashMap index structure is proposed. The secure index is constructed by using pairs coding function and position sensitive hash function, and the K-nearest neighbor algorithm is used to encrypt the index. The similarity between the query keyword vector and the index node is calculated by Euclidian distance measure, and the multi-keyword fuzzy ciphertext search is realized. In addition to eliminating the pre-defined dictionary and reducing the storage overhead, this scheme also realizes the update of the security index without increasing the search complexity. In addition, the combination of Ethereum blockchain technology and searchable encryption scheme avoids data tampering by malicious servers, and the use of smart contracts as trusted third parties for retrieval work can not only prevent keyword guessing attacks within cloud servers, but also solve the problem of incorrect retrieval results. Through security proof analysis, the proposed scheme not only satisfies the semantic security of adaptive keyword selection, but also can protect user privacy and data security. The experimental comparison between this program and other schemes proves that the program has better efficiency in terms of time and cost while ensuring accuracy.
2023, 45(4): 1356-1365.
doi: 10.11999/JEIT220161
Abstract:
Federated learning has the problem of privacy leakage from the gradient. The existing gradient protection schemes based on homomorphic encryption incur a large time cost and the risk of gradient leakage caused by potential collusion between participants and aggregation server. A new federated learning method called FastProtector is proposed, where the idea of SignSGD is introduced when homomorphic encryption is used to protect participant gradients. Exploiting the feature that the majority of positive and negative gradients determine the aggregation result to make the model convergent, the gradient is quantified and the gradient updating mechanism is improved, which can reduce the overhead of gradient encryption. Meanwhile, an additive secret sharing scheme is proposed to protect the gradient ciphertext against collusion attacks between malicious aggregation servers and participants. Experiments on MNIST and CIFAR-10 dataset show that the proposed method can reduce the total encryption and decryption time by about 80% while ensuring high model accuracy.
Federated learning has the problem of privacy leakage from the gradient. The existing gradient protection schemes based on homomorphic encryption incur a large time cost and the risk of gradient leakage caused by potential collusion between participants and aggregation server. A new federated learning method called FastProtector is proposed, where the idea of SignSGD is introduced when homomorphic encryption is used to protect participant gradients. Exploiting the feature that the majority of positive and negative gradients determine the aggregation result to make the model convergent, the gradient is quantified and the gradient updating mechanism is improved, which can reduce the overhead of gradient encryption. Meanwhile, an additive secret sharing scheme is proposed to protect the gradient ciphertext against collusion attacks between malicious aggregation servers and participants. Experiments on MNIST and CIFAR-10 dataset show that the proposed method can reduce the total encryption and decryption time by about 80% while ensuring high model accuracy.
2023, 45(4): 1366-1373.
doi: 10.11999/JEIT220175
Abstract:
In order to optimize the training delay of the hierarchical Federated Learning (FL) global model, focusing on the selfishness of the terminal devices in the actual scene, an incentive mechanism based on game theory is proposed. Under the condition of limited incentive budget, the equilibrium solution between terminal devices and edge servers and the minimum edge model training delay are obtained. Considering the different number of terminal devices, a variable incentive training acceleration algorithm based on Stackelberg game is designed to minimize the training delay of a global model. Simulation results demonstrate that the proposed algorithm can effectively reduce the impact of terminal devices selfishness and improve the training speed of hierarchical federated learning global model.
In order to optimize the training delay of the hierarchical Federated Learning (FL) global model, focusing on the selfishness of the terminal devices in the actual scene, an incentive mechanism based on game theory is proposed. Under the condition of limited incentive budget, the equilibrium solution between terminal devices and edge servers and the minimum edge model training delay are obtained. Considering the different number of terminal devices, a variable incentive training acceleration algorithm based on Stackelberg game is designed to minimize the training delay of a global model. Simulation results demonstrate that the proposed algorithm can effectively reduce the impact of terminal devices selfishness and improve the training speed of hierarchical federated learning global model.
2023, 45(4): 1374-1382.
doi: 10.11999/JEIT220178
Abstract:
It is a difficult problem to perceive comprehensively and accurately the eclipse attack of each node in the blockchain network. For this problem, this paper proposes a blockchain security situational awareness method based on the Markov attack graph and game model. The method combines the characteristics of each node of the blockchain network and the eclipse attack to establish a Markov attack graph model, then quantifies the model to calculate the conversion probability of each attack path, and selects the attack path with higher probability to conduct a multi-stage attack and defense game and calculates the maximum objective function value of both sides. By analyzing these function values, the security situation awareness of the entire blockchain network node is completed, and the purpose of predicting the future security situation and system maintenance is achieved. The experimental comparison shows that the model method not only has a low number of successful intrusions but also has the advantage of ensuring the integrity of the system.
It is a difficult problem to perceive comprehensively and accurately the eclipse attack of each node in the blockchain network. For this problem, this paper proposes a blockchain security situational awareness method based on the Markov attack graph and game model. The method combines the characteristics of each node of the blockchain network and the eclipse attack to establish a Markov attack graph model, then quantifies the model to calculate the conversion probability of each attack path, and selects the attack path with higher probability to conduct a multi-stage attack and defense game and calculates the maximum objective function value of both sides. By analyzing these function values, the security situation awareness of the entire blockchain network node is completed, and the purpose of predicting the future security situation and system maintenance is achieved. The experimental comparison shows that the model method not only has a low number of successful intrusions but also has the advantage of ensuring the integrity of the system.
2023, 45(4): 1383-1390.
doi: 10.11999/JEIT220157
Abstract:
As the most common severe weather, rain can degrade the performance of many vision systems designed for clear imaging conditions. In order to realize the simultaneous removal of rain streaks and rain accumulation, and to deal with various real rain scenes, a two-stage rain image restoration method guided by rain density classification is proposed, which integrates physics model and cGAN refinement. Extensive experiments are conducted on representative synthetic rain datasets and realrain scenes. Quantitative and qualitative results demonstrate the superiority of the proposed method in terms of effectiveness and generalization ability.
As the most common severe weather, rain can degrade the performance of many vision systems designed for clear imaging conditions. In order to realize the simultaneous removal of rain streaks and rain accumulation, and to deal with various real rain scenes, a two-stage rain image restoration method guided by rain density classification is proposed, which integrates physics model and cGAN refinement. Extensive experiments are conducted on representative synthetic rain datasets and realrain scenes. Quantitative and qualitative results demonstrate the superiority of the proposed method in terms of effectiveness and generalization ability.
2023, 45(4): 1391-1400.
doi: 10.11999/JEIT221499
Abstract:
Traditional fusion algorithms of infrared and visible images often have defects such as insufficient target extraction and loss of details, which lead to unsatisfactory fusion effects, and the fused image can not be applied to target detection, tracking or recognition. Therefore, a fusion method of infrared and visible images based on guided filtering and improved maximum Shannon entropy segmentation method using Ant Lion Optimization algorithm (ALO) is proposed. First, Ant Lion Optimized Maximum Entropy Segmentation (ALO-MES) algorithm is used to extract the target from infrared image. Then, the Non-Subsampled Shearlet Transform (NSST) is performed on the infrared and visible images to obtained the low frequency and high frequency sub-bands, and conduct guided filtering for obtained sub-bands. The low-frequency fusion coefficient is obtained from the extracted target image and the enhanced infrared and visible low-frequency components through the fusion rule based on ALO-MES. And the high-frequency fusion coefficient is obtained by the enhanced high-frequency sub-bands components through Dual-Channel Spiking Cortical Model (DCSCM). Finally, the fusion image is obtained by inverse NSST transform. The experimental results show that the proposed algorithm can get fusion image with clear target and background information.
2023, 45(4): 1401-1411.
doi: 10.11999/JEIT220120
Abstract:
There are two major problems in the research on brain functional activation: feature extraction relies on experience; and it is difficult to mine deep physiological information. Focusing on these two problems, this paper proposes a self-adaptive Variational Mode Decomposition (VMD) algorithm by introducing the VMD technique. The algorithm considers the physiological significance of cerebral blood oxygen signals in different frequency bands and reduces the dependence on the selection of hyperparameters. The experimental results show that the self-adaptive VMD algorithm can accurately extract meaningful components in functional Near-InfraRed Spectroscopy (fNIRS), thereby improving the effect of preprocessing. Secondly, this paper proposes Linear Mapping Field (LMF) based on the idea of mapping time series into images and using deep convolutional neural networks for learning. Based on LMF, this paper maps the fNIRS sequence into a two-dimensional image with a low amount of computation supplemented by a deep convolutional neural network, and realizes the extraction of deep features of fNIRS physiological signals. The experimental results demonstrate the performance and advantages of the proposed LMF. Finally, this paper discusses and analyzes the effectiveness of the proposed methods, indicating that different from recurrent neural networks which can only perceive the time series in a “sequential” manner, the convolutional neural networks’ characteristic of "jumping" perception is the key to achieving excellent results.
There are two major problems in the research on brain functional activation: feature extraction relies on experience; and it is difficult to mine deep physiological information. Focusing on these two problems, this paper proposes a self-adaptive Variational Mode Decomposition (VMD) algorithm by introducing the VMD technique. The algorithm considers the physiological significance of cerebral blood oxygen signals in different frequency bands and reduces the dependence on the selection of hyperparameters. The experimental results show that the self-adaptive VMD algorithm can accurately extract meaningful components in functional Near-InfraRed Spectroscopy (fNIRS), thereby improving the effect of preprocessing. Secondly, this paper proposes Linear Mapping Field (LMF) based on the idea of mapping time series into images and using deep convolutional neural networks for learning. Based on LMF, this paper maps the fNIRS sequence into a two-dimensional image with a low amount of computation supplemented by a deep convolutional neural network, and realizes the extraction of deep features of fNIRS physiological signals. The experimental results demonstrate the performance and advantages of the proposed LMF. Finally, this paper discusses and analyzes the effectiveness of the proposed methods, indicating that different from recurrent neural networks which can only perceive the time series in a “sequential” manner, the convolutional neural networks’ characteristic of "jumping" perception is the key to achieving excellent results.
2023, 45(4): 1412-1425.
doi: 10.11999/JEIT220172
Abstract:
In recent years, medical image processing with CNN has made remarkable research progress in the task of chest film disease classification. However, compared with single structure CNN, dual-path network can combine the characteristics of different CNN to improve the ability of disease classification. Secondly, for different diseases, their location, size, shape, and texture are different, the attention mechanism helps the model to extract different pathological features and improve the classification accuracy. Therefore, focusing on the chest film disease classification problem, a dual path convolution neural network TADPN(Triple Attention Dual Path Network) combined with a triple attention mechanism is proposed. TADPN takes the dual-path network combined with ResNet and DenseNet as the backbone network and uses three different forms of attention mechanisms to improve the backbone network. The network complexity and classification accuracy are improved while maintaining the stability of the parameters. In this paper, the validity of TADPN is compared with the six advanced algorithms on the ChestXray14 dataset. The experiments show the progressiveness of the dual-path CNN and the triple attention mechanism, as well as the effectiveness of TADPN. The average AUC value of 14 diseases reaches 0.8185, which is 1.1% higher than that of previous generations.
In recent years, medical image processing with CNN has made remarkable research progress in the task of chest film disease classification. However, compared with single structure CNN, dual-path network can combine the characteristics of different CNN to improve the ability of disease classification. Secondly, for different diseases, their location, size, shape, and texture are different, the attention mechanism helps the model to extract different pathological features and improve the classification accuracy. Therefore, focusing on the chest film disease classification problem, a dual path convolution neural network TADPN(Triple Attention Dual Path Network) combined with a triple attention mechanism is proposed. TADPN takes the dual-path network combined with ResNet and DenseNet as the backbone network and uses three different forms of attention mechanisms to improve the backbone network. The network complexity and classification accuracy are improved while maintaining the stability of the parameters. In this paper, the validity of TADPN is compared with the six advanced algorithms on the ChestXray14 dataset. The experiments show the progressiveness of the dual-path CNN and the triple attention mechanism, as well as the effectiveness of TADPN. The average AUC value of 14 diseases reaches 0.8185, which is 1.1% higher than that of previous generations.
2023, 45(4): 1426-1434.
doi: 10.11999/JEIT220204
Abstract:
In recent years, graph convolutional network has been widely used in hyperspectral image classification because of its feature aggregation mechanism, which can simultaneously represent the features of a single node and neighboring nodes. However, there are many problems in HyperSpectral Images(HSI), such as band redundancy and different spectrum of the same object, which results in the inadequate reliability of the initial graph constructed by directly using the original spectral features, thus leading to the low classification accuracy of hyperspectral images. Therefore, a semi-supervised classification method for hyperspectral images based on Spectral Attention Graph Convolutional Network (SAGCN) is proposed. Firstly, the attention module is used to interact with the local and global information of the spectrum, and realize the adaptive weighting of the spectrum. Then, for the hyperspectral images after spectral weighting, a more accurate nearest neighbor matrix is constructed by using spatial-spectral similarity. Finally, effective feature aggregation of labeled and unlabeled samples is carried out by graph convolution, and the network is trained with the features of labeled samples. Experimental results on three real hyperspectral image datasets including Indian Pines, Kennedy Space Center and Botswana demonstrate the effectiveness of the proposed method.
In recent years, graph convolutional network has been widely used in hyperspectral image classification because of its feature aggregation mechanism, which can simultaneously represent the features of a single node and neighboring nodes. However, there are many problems in HyperSpectral Images(HSI), such as band redundancy and different spectrum of the same object, which results in the inadequate reliability of the initial graph constructed by directly using the original spectral features, thus leading to the low classification accuracy of hyperspectral images. Therefore, a semi-supervised classification method for hyperspectral images based on Spectral Attention Graph Convolutional Network (SAGCN) is proposed. Firstly, the attention module is used to interact with the local and global information of the spectrum, and realize the adaptive weighting of the spectrum. Then, for the hyperspectral images after spectral weighting, a more accurate nearest neighbor matrix is constructed by using spatial-spectral similarity. Finally, effective feature aggregation of labeled and unlabeled samples is carried out by graph convolution, and the network is trained with the features of labeled samples. Experimental results on three real hyperspectral image datasets including Indian Pines, Kennedy Space Center and Botswana demonstrate the effectiveness of the proposed method.
2023, 45(4): 1435-1444.
doi: 10.11999/JEIT220201
Abstract:
Most state-of-the-art building extraction from satellite imagery are based on binary segmentation. However, the geographic information has not been considered in these methods, thus, it is difficult to extract building accurately. To consider fully the geographic information on feature extraction, a building extraction convolutional neural network based on footprint map and bidirectional connection is proposed. The proposed method is a multi-branch network, which is designed to predict the footprint and bidirectional connection map, respectively. This paper predicts the footprint heatmap of buildings and uses the Non-Maximum Suppression (NMS) algorithm to obtain the pixel coordinates. Another two branches are used to predict positive connectivity and negative connectivity between footprints. Each pair of nodes is connected according to the bidirectional connectivity map to obtain the final building outline. Experiments on the Buildings2Vec dataset demonstrate that the proposed method outperforms various previous work, which illustrate the superiority in building extraction from satellite imagery.
Most state-of-the-art building extraction from satellite imagery are based on binary segmentation. However, the geographic information has not been considered in these methods, thus, it is difficult to extract building accurately. To consider fully the geographic information on feature extraction, a building extraction convolutional neural network based on footprint map and bidirectional connection is proposed. The proposed method is a multi-branch network, which is designed to predict the footprint and bidirectional connection map, respectively. This paper predicts the footprint heatmap of buildings and uses the Non-Maximum Suppression (NMS) algorithm to obtain the pixel coordinates. Another two branches are used to predict positive connectivity and negative connectivity between footprints. Each pair of nodes is connected according to the bidirectional connectivity map to obtain the final building outline. Experiments on the Buildings2Vec dataset demonstrate that the proposed method outperforms various previous work, which illustrate the superiority in building extraction from satellite imagery.
2023, 45(4): 1445-1453.
doi: 10.11999/JEIT220122
Abstract:
In pedestrian detection, small-scale pedestrians are often missed and mistakenly detected. In order to improve detection precision and reduce miss detection rate of small-scale pedestrians, a feature enhancement module is proposed. First, considering the problem that small-scale pedestrians feature gradually decreases as network goes deeper, feature fusion strategy breaks through the constraints of feature pyramid structure and fuses deep and shallow feature maps to retain lots of small-scale pedestrian features. Then, considering the problem that small-scale pedestrian features are easily confused with background information, self-attention module combined with channel attention module models the spatial and channel correlation of feature maps, using small-scale pedestrian contextual information and channel information to enhance small-scale pedestrian features and suppress background information. Finally, a small-scale pedestrian detector is constructed based on the feature enhancement module. For small-scale pedestrians, the proposed algorithm has 19.8% detection accuracy, 22 frames per second speed on CrowdHuman dataset and 13.1% miss rate on CityPersons dataset. The results show that the proposed algorithm performs better than other compared algorithms for small-scale pedestrian detection and achieves faster detection speed.
In pedestrian detection, small-scale pedestrians are often missed and mistakenly detected. In order to improve detection precision and reduce miss detection rate of small-scale pedestrians, a feature enhancement module is proposed. First, considering the problem that small-scale pedestrians feature gradually decreases as network goes deeper, feature fusion strategy breaks through the constraints of feature pyramid structure and fuses deep and shallow feature maps to retain lots of small-scale pedestrian features. Then, considering the problem that small-scale pedestrian features are easily confused with background information, self-attention module combined with channel attention module models the spatial and channel correlation of feature maps, using small-scale pedestrian contextual information and channel information to enhance small-scale pedestrian features and suppress background information. Finally, a small-scale pedestrian detector is constructed based on the feature enhancement module. For small-scale pedestrians, the proposed algorithm has 19.8% detection accuracy, 22 frames per second speed on CrowdHuman dataset and 13.1% miss rate on CityPersons dataset. The results show that the proposed algorithm performs better than other compared algorithms for small-scale pedestrian detection and achieves faster detection speed.
2023, 45(4): 1454-1463.
doi: 10.11999/JEIT220113
Abstract:
Deep learning algorithms represented by Convolutional Neural Networks (CNN) are highly dependent on the nonlinearity of the model and debugging techniques, which have generally black-box properties during practical applications, limiting severely their further development in security-sensitive fields. To this end, a Coarse-to-Fine Class Activation Mapping (CF-CAM) algorithm is proposed for diagnosing the decision-making behaviors of deep neural networks. The algorithm re-establishes the relationship between the feature map and the model decision, uses the contrastive layer-wise relevance propagation theory to obtain the contribution of each position in the feature map to the network decision, generates a spatial-level correlation mask and finds the important area that affects the model decision. After that, the mask is linearly weighted with the fuzzed input image and re-input into the network to obtain the target score of the feature map, and the deep neural network is explained from the coarse stage to the fine stage in the spatial domain and the channel domain. The experimental results show that the CF-CAM proposed in this paper has obvious advantages in terms of faithfulness and localization performance compared to other methods. In addition, this paper applies CF-CAM as a data enhancement strategy for the task of fine-grained classification of birds, which can effectively improve the accuracy of network recognition by learning difficult samples, further verify the effectiveness and superiority of this method.
Deep learning algorithms represented by Convolutional Neural Networks (CNN) are highly dependent on the nonlinearity of the model and debugging techniques, which have generally black-box properties during practical applications, limiting severely their further development in security-sensitive fields. To this end, a Coarse-to-Fine Class Activation Mapping (CF-CAM) algorithm is proposed for diagnosing the decision-making behaviors of deep neural networks. The algorithm re-establishes the relationship between the feature map and the model decision, uses the contrastive layer-wise relevance propagation theory to obtain the contribution of each position in the feature map to the network decision, generates a spatial-level correlation mask and finds the important area that affects the model decision. After that, the mask is linearly weighted with the fuzzed input image and re-input into the network to obtain the target score of the feature map, and the deep neural network is explained from the coarse stage to the fine stage in the spatial domain and the channel domain. The experimental results show that the CF-CAM proposed in this paper has obvious advantages in terms of faithfulness and localization performance compared to other methods. In addition, this paper applies CF-CAM as a data enhancement strategy for the task of fine-grained classification of birds, which can effectively improve the accuracy of network recognition by learning difficult samples, further verify the effectiveness and superiority of this method.
2023, 45(4): 1464-1474.
doi: 10.11999/JEIT220217
Abstract:
Considering the low efficiency of traditional ElectroCardioGram(ECG) classification methods, an accurate and fast ElectroCardioGram classification method based on Adaptive Fast S-Transform (AFST) and XGBoost is proposed. Firstly, the main feature points of the ECG signals are determined through a fast positioning algorithm, and then the S-Transform window width factor is adjusted adaptively according to the main feature points to enhance the time-frequency resolution of the S-transform while avoiding iterative calculation and reducing the running time greatly; Secondly, based on the time-frequency matrix of AFST, 12 eigenvalues are extracted to represent the characteristic information of 5 kinds of ECG signals, with low eigenvector dimension and strong recognition ability. Finally, XGBoost is used to identify the eigenvectors. The experimental studies based on the MIT-BIH arrhythmia database and the verification of patient measurement data show that, with the proposed method, the classification time of ECG signals is significantly shortened and classification accuracy of 99.59%, 97.32% is obtained respectively, which is suitable for the rapid diagnosis of abnormal diseases in the center rate of the actual detection system.
Considering the low efficiency of traditional ElectroCardioGram(ECG) classification methods, an accurate and fast ElectroCardioGram classification method based on Adaptive Fast S-Transform (AFST) and XGBoost is proposed. Firstly, the main feature points of the ECG signals are determined through a fast positioning algorithm, and then the S-Transform window width factor is adjusted adaptively according to the main feature points to enhance the time-frequency resolution of the S-transform while avoiding iterative calculation and reducing the running time greatly; Secondly, based on the time-frequency matrix of AFST, 12 eigenvalues are extracted to represent the characteristic information of 5 kinds of ECG signals, with low eigenvector dimension and strong recognition ability. Finally, XGBoost is used to identify the eigenvectors. The experimental studies based on the MIT-BIH arrhythmia database and the verification of patient measurement data show that, with the proposed method, the classification time of ECG signals is significantly shortened and classification accuracy of 99.59%, 97.32% is obtained respectively, which is suitable for the rapid diagnosis of abnormal diseases in the center rate of the actual detection system.
2023, 45(4): 1475-1484.
doi: 10.11999/JEIT220218
Abstract:
Understanding the attention mechanism of the human visual system has attracted much research attention from researchers and industries. Recent studies of attention mechanisms focus mainly on observer patterns. However, more intelligent applications are presented in the real world and require objective visual attention detection. Automating tasks such as surveillance or human-robot collaboration require anticipating and predicting the behavior of objects. In such contexts, gaze and focus can be highly informative about participants' intentions, goals, and upcoming decisions. Here, a progressive mechanism of objective visual attention is developed by combining cognitive mechanisms. The field is first viewed as a combination of geometric structure and geometric details. A Hierarchical Self-Attention Module (HSAM) is constructed to capture the long-distance dependencies between deep features and adapt geometric feature diversity. With the identified generators, the field of view direction vectors are generated, and the probability distribution of gaze points is obtained. Furthermore, a feature fusion module is designed for structure sharing, fusion, and enhancement of multi-resolution features. Its output contains more detailed spatial and global information, better obtaining spatial context features. The experimental results are in excellent agreement with theoretical predictions by different evaluation metrics for objective attention estimation on publicly available and self-built datasets.
Understanding the attention mechanism of the human visual system has attracted much research attention from researchers and industries. Recent studies of attention mechanisms focus mainly on observer patterns. However, more intelligent applications are presented in the real world and require objective visual attention detection. Automating tasks such as surveillance or human-robot collaboration require anticipating and predicting the behavior of objects. In such contexts, gaze and focus can be highly informative about participants' intentions, goals, and upcoming decisions. Here, a progressive mechanism of objective visual attention is developed by combining cognitive mechanisms. The field is first viewed as a combination of geometric structure and geometric details. A Hierarchical Self-Attention Module (HSAM) is constructed to capture the long-distance dependencies between deep features and adapt geometric feature diversity. With the identified generators, the field of view direction vectors are generated, and the probability distribution of gaze points is obtained. Furthermore, a feature fusion module is designed for structure sharing, fusion, and enhancement of multi-resolution features. Its output contains more detailed spatial and global information, better obtaining spatial context features. The experimental results are in excellent agreement with theoretical predictions by different evaluation metrics for objective attention estimation on publicly available and self-built datasets.
2023, 45(4): 1485-1493.
doi: 10.11999/JEIT220270
Abstract:
In recent years, skeleton-based human action recognition has attracted widespread attention because of the robustness and generalization ability of skeleton data. Among them, the graph convolutional network that models the human skeleton into a spatiotemporal graph has achieved remarkable performance. However, graph convolutions learn mainly long-term interactive connections through a series of 3D convolutions, which are localized and limited by the size of convolution kernels, which can not effectively capture long-range dependencies. In this paper, a Collaborative Convolutional Transformer (Co-ConvT) network is proposed to establish remote dependencies by introducing Transformer's self-attention mechanism and combining it with Graph Convolutional Neural Networks (GCNs) for action recognition, enabling the model to extract local information through graph convolution while capturing the rich remote dependencies through Transformer. In addition, Transformer's self-attention mechanism is calculated at the pixel level, a huge computational cost is generated. The model divides the entire network into two stages. The first stage uses pure convolution to extract shallow spatial features, and the second stage uses the proposed ConvT block to capture high-level semantic information, reducing the computational complexity. Moreover, the linear embeddings in the original Transformer are replaced with convolutional embeddings to obtain local spatial information enhancement, and thus removing the positional encoding in the original model, making the model lighter. Experimentally validated on two large-scale authoritative datasets NTU-RGB+D and Kinetics-Skeleton, the model achieves respectively Top-1 accuracy of 88.1% and 36.6%. The experimental results demonstrate that the performance of the model is greatly improved.
In recent years, skeleton-based human action recognition has attracted widespread attention because of the robustness and generalization ability of skeleton data. Among them, the graph convolutional network that models the human skeleton into a spatiotemporal graph has achieved remarkable performance. However, graph convolutions learn mainly long-term interactive connections through a series of 3D convolutions, which are localized and limited by the size of convolution kernels, which can not effectively capture long-range dependencies. In this paper, a Collaborative Convolutional Transformer (Co-ConvT) network is proposed to establish remote dependencies by introducing Transformer's self-attention mechanism and combining it with Graph Convolutional Neural Networks (GCNs) for action recognition, enabling the model to extract local information through graph convolution while capturing the rich remote dependencies through Transformer. In addition, Transformer's self-attention mechanism is calculated at the pixel level, a huge computational cost is generated. The model divides the entire network into two stages. The first stage uses pure convolution to extract shallow spatial features, and the second stage uses the proposed ConvT block to capture high-level semantic information, reducing the computational complexity. Moreover, the linear embeddings in the original Transformer are replaced with convolutional embeddings to obtain local spatial information enhancement, and thus removing the positional encoding in the original model, making the model lighter. Experimentally validated on two large-scale authoritative datasets NTU-RGB+D and Kinetics-Skeleton, the model achieves respectively Top-1 accuracy of 88.1% and 36.6%. The experimental results demonstrate that the performance of the model is greatly improved.
2023, 45(4): 1494-1504.
doi: 10.11999/JEIT220182
Abstract:
The unsupervised image segmentation method is sensitive to noise, leading to difficult building image model and poor accuracy of segmentation results. In this paper, a minimum spanning tree segmentation and extract with image edge weight optimization is proposed. Firstly, L0 gradient minimum is used to smooth the noise. The Canny edge detection with Otsu is optimized to obtain more accurate edge information. Secondly, the weight function is redesigned and the weighted graph by using more reasonable color difference space is constructed. The segmentation criterion is improved to optimize the process of object merging and distinguishing. Finally, different types of images are chosen to conduct experiments with noise resistance and segmentation effect. Experimental comparing results show that the proposed algorithm has excellent anti-noise performance, and the segmentation accuracy is improved by 5.15% on average, the over-segmentation rate is decreased by 32.07% on average, and the under-segmentation rate is decreased by 2.69% on average. Moreover, this method is applied to the river and lake extraction of aviation and remote sensing images, and the result has more complete structure, less irrelevant information and better anti-noise performance.
The unsupervised image segmentation method is sensitive to noise, leading to difficult building image model and poor accuracy of segmentation results. In this paper, a minimum spanning tree segmentation and extract with image edge weight optimization is proposed. Firstly, L0 gradient minimum is used to smooth the noise. The Canny edge detection with Otsu is optimized to obtain more accurate edge information. Secondly, the weight function is redesigned and the weighted graph by using more reasonable color difference space is constructed. The segmentation criterion is improved to optimize the process of object merging and distinguishing. Finally, different types of images are chosen to conduct experiments with noise resistance and segmentation effect. Experimental comparing results show that the proposed algorithm has excellent anti-noise performance, and the segmentation accuracy is improved by 5.15% on average, the over-segmentation rate is decreased by 32.07% on average, and the under-segmentation rate is decreased by 2.69% on average. Moreover, this method is applied to the river and lake extraction of aviation and remote sensing images, and the result has more complete structure, less irrelevant information and better anti-noise performance.
2023, 45(4): 1505-1513.
doi: 10.11999/JEIT220249
Abstract:
Mobile Crowd Sensing (MCS) is a new paradigm that collects sensing data via the mobility of massive workers and carried sensing devices. Current works focus on the task allocation issue and improving sensing data quality. However, they ignore the sensing tasks lacking qualified workers and decrease the task completion quality. To tackle the above problem, for sensing tasks that lack qualified workers, inexperienced workers are developed to qualified workers and execute these tasks. As a result, the qualified workers can long-term execute these tasks, and the sensing data quality and long-term platform utility are improved. Furthermore, both the capacities that sensing tasks need and the capacities that workers own are considered. According the above capacities, first, a stable matching algorithm is applied to select workers to be developed. And then a Multi-stage Worker Selection and Development (MWSD) algorithm is proposed based on capacity fusion and Semi-Markov prediction. The results show that compared to Blockchain-based Nondeterministic Teamwork Cooperation (BNTC) algorithm, the mechanism can improve the data quality of sensing tasks lacking qualified workers by 24% and long-term platform utility by 17%.
Mobile Crowd Sensing (MCS) is a new paradigm that collects sensing data via the mobility of massive workers and carried sensing devices. Current works focus on the task allocation issue and improving sensing data quality. However, they ignore the sensing tasks lacking qualified workers and decrease the task completion quality. To tackle the above problem, for sensing tasks that lack qualified workers, inexperienced workers are developed to qualified workers and execute these tasks. As a result, the qualified workers can long-term execute these tasks, and the sensing data quality and long-term platform utility are improved. Furthermore, both the capacities that sensing tasks need and the capacities that workers own are considered. According the above capacities, first, a stable matching algorithm is applied to select workers to be developed. And then a Multi-stage Worker Selection and Development (MWSD) algorithm is proposed based on capacity fusion and Semi-Markov prediction. The results show that compared to Blockchain-based Nondeterministic Teamwork Cooperation (BNTC) algorithm, the mechanism can improve the data quality of sensing tasks lacking qualified workers by 24% and long-term platform utility by 17%.
2023, 45(4): 1514-1528.
doi: 10.11999/JEIT220116
Abstract:
As an emerging technology, edge intelligence is receiving extensive attention from scholars at home and abroad. As a combination of artificial intelligence technology and edge computing technology, it is expected to promote the deployment of artificial intelligence technology in various industries and accelerate the process of industrial intelligence. In this paper the basic principles, system architecture, and comparative advantages of edge intelligence technology, and sorts out the research status of edge intelligence technology at home and abroad are first introduced. The application prospects of the life cycle, the application of edge intelligence technology in the whole life cycle of rail transit process management and control, construction site data collection and analysis, information sharing, intelligent operation and maintenance, intelligent scheduling, automatic driving system, train coordination control, and transformation and upgrading are described in detail. Then the designs and implements an edge intelligent platform under the background of rail transit intelligent operation control, and tests the functions and performance of edge intelligence applications based on deep learning and reinforcement learning are discussed. Finally, the problems and challenges in the application of edge intelligence technology to the field of rail transit are summarized. The research in this paper is expected to provide a useful reference and practical basis for edge intelligence applications to the field of rail transit.
As an emerging technology, edge intelligence is receiving extensive attention from scholars at home and abroad. As a combination of artificial intelligence technology and edge computing technology, it is expected to promote the deployment of artificial intelligence technology in various industries and accelerate the process of industrial intelligence. In this paper the basic principles, system architecture, and comparative advantages of edge intelligence technology, and sorts out the research status of edge intelligence technology at home and abroad are first introduced. The application prospects of the life cycle, the application of edge intelligence technology in the whole life cycle of rail transit process management and control, construction site data collection and analysis, information sharing, intelligent operation and maintenance, intelligent scheduling, automatic driving system, train coordination control, and transformation and upgrading are described in detail. Then the designs and implements an edge intelligent platform under the background of rail transit intelligent operation control, and tests the functions and performance of edge intelligence applications based on deep learning and reinforcement learning are discussed. Finally, the problems and challenges in the application of edge intelligence technology to the field of rail transit are summarized. The research in this paper is expected to provide a useful reference and practical basis for edge intelligence applications to the field of rail transit.
