Email alert
2019 Vol. 41, No. 10
Display Method:
2019, 41(10): 2287-2293.
doi: 10.11999/JEIT190149
Abstract:
Considering the resource allocation problem for Device-to-Device (D2D) communications, a channel selection and power control strategy for D2D communications is investigated. On the premise of guaranteeing the Quality of Service (QoS) of cellular users, a heuristic based D2D channel selection algorithm is proposed to find the suitable channel reusing resources for D2D users in the system. At the same time, the optimal transmission power of D2D users is obtained by using the Lagrange dual method. Simulation results demonstrate that when the cellular user shares channel resources with multiple pairs of D2D users, the system throughput can be dramatically improved. The performance of this algorithm outperforms the exiting algorithms under the same conditions.
Considering the resource allocation problem for Device-to-Device (D2D) communications, a channel selection and power control strategy for D2D communications is investigated. On the premise of guaranteeing the Quality of Service (QoS) of cellular users, a heuristic based D2D channel selection algorithm is proposed to find the suitable channel reusing resources for D2D users in the system. At the same time, the optimal transmission power of D2D users is obtained by using the Lagrange dual method. Simulation results demonstrate that when the cellular user shares channel resources with multiple pairs of D2D users, the system throughput can be dramatically improved. The performance of this algorithm outperforms the exiting algorithms under the same conditions.
2019, 41(10): 2294-2301.
doi: 10.11999/JEIT190079
Abstract:
When the Underwater Wireless Sensor Network (UWSN) performs target tracking, the contributions of the measured values of the nodes are different, and the battery energy carried by the sensor node is limited. Therefore, a good node fusion weight method and node planning mechanism can obtain better tracking performance. A distributed particle filter target tracking algorithm based on Grubbs criterion and Mutual Information Entropy Weighted (GMIEW) fusion is proposed to solve the above problem in this paper. Firstly, the Grubbs criterion is used to analyze and verify the information obtained by the sensor nodes before the information fusion, and the interference information and error information are removed. Secondly, in the process of calculating the importance weight of particle filter, the dynamic weighting factor is introduced. The mutual information entropy between the measured value of the sensor node and the target state is used to reflect the amount of target information provided by the sensor node, so as to obtain the corresponding weighting factor of each node. Finally, the improved cluster-tree network topology is used to track the target in three-dimensional space. Simulation results show that the proposed algorithm improves greatly the accuracy of underwater sensor measurement data for target tracking prediction and reduces the tracking error.
When the Underwater Wireless Sensor Network (UWSN) performs target tracking, the contributions of the measured values of the nodes are different, and the battery energy carried by the sensor node is limited. Therefore, a good node fusion weight method and node planning mechanism can obtain better tracking performance. A distributed particle filter target tracking algorithm based on Grubbs criterion and Mutual Information Entropy Weighted (GMIEW) fusion is proposed to solve the above problem in this paper. Firstly, the Grubbs criterion is used to analyze and verify the information obtained by the sensor nodes before the information fusion, and the interference information and error information are removed. Secondly, in the process of calculating the importance weight of particle filter, the dynamic weighting factor is introduced. The mutual information entropy between the measured value of the sensor node and the target state is used to reflect the amount of target information provided by the sensor node, so as to obtain the corresponding weighting factor of each node. Finally, the improved cluster-tree network topology is used to track the target in three-dimensional space. Simulation results show that the proposed algorithm improves greatly the accuracy of underwater sensor measurement data for target tracking prediction and reduces the tracking error.
2019, 41(10): 2302-2309.
doi: 10.11999/JEIT180599
Abstract:
The radio map construction is time consuming and labor intensive, and the conventional semi-supervised based methods usually ignore the influence of the uneven distribution of high-dimensional Received Signal Strength (RSS). In order to solve that problem, a semi-supervised radio map construction approach which is based on the nonhomogeneous distribution characteristic of RSS is proposed. The approach utilizes the RSS local scale and common neighbors similarities to calculate the weighted matrix. Thus, the weighted graph that reflects accurately the structure of RSS data manifold is presented. In addition, the weighted graph is used to find the optimal solution of the objective function to calibrate the locations of plenty of unlabeled data by a small number of labeled RSS. The extensive experiments demonstrate that the proposed method is capable of not only constructing an accurate radio map at a low manual cost, but also achieving a high localization accuracy.
The radio map construction is time consuming and labor intensive, and the conventional semi-supervised based methods usually ignore the influence of the uneven distribution of high-dimensional Received Signal Strength (RSS). In order to solve that problem, a semi-supervised radio map construction approach which is based on the nonhomogeneous distribution characteristic of RSS is proposed. The approach utilizes the RSS local scale and common neighbors similarities to calculate the weighted matrix. Thus, the weighted graph that reflects accurately the structure of RSS data manifold is presented. In addition, the weighted graph is used to find the optimal solution of the objective function to calibrate the locations of plenty of unlabeled data by a small number of labeled RSS. The extensive experiments demonstrate that the proposed method is capable of not only constructing an accurate radio map at a low manual cost, but also achieving a high localization accuracy.
2019, 41(10): 2310-2317.
doi: 10.11999/JEIT190094
Abstract:
Considering coverage redundancy problem existed in random heterogeneous sensor networks with high density deployment, a Node Scheduling algorithm for Stochastic Heterogeneous wireless sensor networks(NSSH) is proposed. The Delaunary triangulation is constructed based on the network prototype topology to work out a local subset of nodes for localization scheduling. Independent configuration of the perceived radius is achieved by discounting the radius of the circumcircle with the adjacent node. The concept of geometric line and plane is introduced, and the overlapping area and the effective constrained arcs are used to classify and identify the grey and black nodes. So the node only relies on local and neighbor information for radius adjustment and redundant node sleep. The simulation results show that NSSH can approximately match the dropping redundancy of greedy algorithm at the cost of low complexity, and exhibit low sensitivity to network size, heterogeneous span and parameter configuration.
Considering coverage redundancy problem existed in random heterogeneous sensor networks with high density deployment, a Node Scheduling algorithm for Stochastic Heterogeneous wireless sensor networks(NSSH) is proposed. The Delaunary triangulation is constructed based on the network prototype topology to work out a local subset of nodes for localization scheduling. Independent configuration of the perceived radius is achieved by discounting the radius of the circumcircle with the adjacent node. The concept of geometric line and plane is introduced, and the overlapping area and the effective constrained arcs are used to classify and identify the grey and black nodes. So the node only relies on local and neighbor information for radius adjustment and redundant node sleep. The simulation results show that NSSH can approximately match the dropping redundancy of greedy algorithm at the cost of low complexity, and exhibit low sensitivity to network size, heterogeneous span and parameter configuration.
2019, 41(10): 2318-2324.
doi: 10.11999/JEIT181181
Abstract:
Considering the problem of Orthogonal Frequency Division Multiplexing (OFDM) signal delay estimation with only a Single Measurement Vector (SMV) in a complex environment, a sparse reconstruction time delay estimation algorithm based on Bayesian Automatic Relevance Determination (BARD) is proposed. The Bayesian framework is used to start from the perspective of further mining useful information, and asymmetric Automatic Relevance Determination(ARD) priori is introduced to integrate into the parameter estimation process, which improves the accuracy of time delay estimation under SMV and low Signal-to-Noise Ratio (SNR) conditions. Firstly, a sparse real-domain representation model is constructed based on the estimated frequency domain response of the OFDM signal physical layer protocol data unit. Then, probability hypothesis for the noise and sparse coefficient vectors are made in the model, and Automatic Relevance Determination (ARD) prior is introduced. Finally, according to the Bayesian framework, the Expectation Maximization (EM) algorithm is used to solve the hyperparameters to estimate the delay. The simulation experiments show that the proposed algorithm has better estimation performance and is closer to the Cramér–Rao Bound (CRB). At the same time, based on the Universal Software Radio Peripheral (USRP), the effectiveness of the proposed algorithm is verified by the actual signal.
Considering the problem of Orthogonal Frequency Division Multiplexing (OFDM) signal delay estimation with only a Single Measurement Vector (SMV) in a complex environment, a sparse reconstruction time delay estimation algorithm based on Bayesian Automatic Relevance Determination (BARD) is proposed. The Bayesian framework is used to start from the perspective of further mining useful information, and asymmetric Automatic Relevance Determination(ARD) priori is introduced to integrate into the parameter estimation process, which improves the accuracy of time delay estimation under SMV and low Signal-to-Noise Ratio (SNR) conditions. Firstly, a sparse real-domain representation model is constructed based on the estimated frequency domain response of the OFDM signal physical layer protocol data unit. Then, probability hypothesis for the noise and sparse coefficient vectors are made in the model, and Automatic Relevance Determination (ARD) prior is introduced. Finally, according to the Bayesian framework, the Expectation Maximization (EM) algorithm is used to solve the hyperparameters to estimate the delay. The simulation experiments show that the proposed algorithm has better estimation performance and is closer to the Cramér–Rao Bound (CRB). At the same time, based on the Universal Software Radio Peripheral (USRP), the effectiveness of the proposed algorithm is verified by the actual signal.
2019, 41(10): 2325-2333.
doi: 10.11999/JEIT190069
Abstract:
To solve the problems of high encoding complexity and long encoding delay in the multi-source multi-relay Low Density Parity Check (LDPC) coded cooperative system, a special kind of structured LDPC codes—Quasi-Cyclic LDPC (QC-LDPC) codes based on generator matrix is proposed, which combines the characteristics of QC-LDPC codes and Generator-matrix-based LDPC (G-LDPC) codes. It can perform completely parallel encoding, which greatly reduces the encoding complexity and delay at the relays. Based on this, a joint parity check matrix corresponding to the QC-LDPC codes adopted by the sources and relays is deduced, and the matrix is further jointly designed based on the Greatest Common Divisor (GCD) theorem to eliminate all cycles of girth-4 and girth-6. Theoretical analysis and simulation results show that under the same conditions, the Bit Error Rate (BER) performance of the proposed system is better than that of the corresponding point-to-point system. The simulation results also show that the cooperative system with jointly designed QC-LDPC codes can obtain a higher coding gain than the system with explicitly constructed QC-LDPC codes or generally constructed QC-LDPC codes.
To solve the problems of high encoding complexity and long encoding delay in the multi-source multi-relay Low Density Parity Check (LDPC) coded cooperative system, a special kind of structured LDPC codes—Quasi-Cyclic LDPC (QC-LDPC) codes based on generator matrix is proposed, which combines the characteristics of QC-LDPC codes and Generator-matrix-based LDPC (G-LDPC) codes. It can perform completely parallel encoding, which greatly reduces the encoding complexity and delay at the relays. Based on this, a joint parity check matrix corresponding to the QC-LDPC codes adopted by the sources and relays is deduced, and the matrix is further jointly designed based on the Greatest Common Divisor (GCD) theorem to eliminate all cycles of girth-4 and girth-6. Theoretical analysis and simulation results show that under the same conditions, the Bit Error Rate (BER) performance of the proposed system is better than that of the corresponding point-to-point system. The simulation results also show that the cooperative system with jointly designed QC-LDPC codes can obtain a higher coding gain than the system with explicitly constructed QC-LDPC codes or generally constructed QC-LDPC codes.
2019, 41(10): 2334-2341.
doi: 10.11999/JEIT190034
Abstract:
Two online blind equalization algorithms based on Echo State Network (ESN) in this paper are proposed for the nonlinear satellite channel. These two algorithms take advantage of the good nonlinear approximation of ESN to bring the High-Order Statistics (HOS) of the transmitted signal into the ESN, and constructing cost function of blind equalization by combining Constant Modulus Algorithm (CMA) and Multi-Modulus Algorithm (MMA). Then, the Recursive Least Squares (RLS) algorithm is used to iteratively optimize the network output weights, and the online blind equalization of the constant modulus signals and the multi-modulus signals over the channel of Volterra satellite are realized. Experiments show that the proposed algorithms can effectively reduce the distortion of the transmitted signal by the nonlinear channel. Compared with the traditional Volterra filtering method, they have faster convergence speed and lower mean square error.
Two online blind equalization algorithms based on Echo State Network (ESN) in this paper are proposed for the nonlinear satellite channel. These two algorithms take advantage of the good nonlinear approximation of ESN to bring the High-Order Statistics (HOS) of the transmitted signal into the ESN, and constructing cost function of blind equalization by combining Constant Modulus Algorithm (CMA) and Multi-Modulus Algorithm (MMA). Then, the Recursive Least Squares (RLS) algorithm is used to iteratively optimize the network output weights, and the online blind equalization of the constant modulus signals and the multi-modulus signals over the channel of Volterra satellite are realized. Experiments show that the proposed algorithms can effectively reduce the distortion of the transmitted signal by the nonlinear channel. Compared with the traditional Volterra filtering method, they have faster convergence speed and lower mean square error.
2019, 41(10): 2342-2349.
doi: 10.11999/JEIT190157
Abstract:
Considering the problem that the Dual-Sequence Frequency Hopping (DSFH) can not communicate at extremely low Signal-to-Noise Ratio (SNR), a Stochastic Resonance (SR) detection method is proposed. The SR takes full advantage of the physical characteristics of DSFH signal to improve the detection performance. Firstly, the SR is constructed by analyzing signals of transmission, reception and the Intermediate Frequency (IF). The scale transaction is used to adjust the IF signal to fit the SR. Secondly, the non-autonomous Fokker-Plank Equation (FPE) is transformed into an autonomous equation by introducing the decision time. Therefore, the analytical solution of the probability density function with the parameter of decision time is obtained. Finally, the detection probability, false alarm probability and Receiver Operating Characteristics (ROC) curve are obtained, when the criterion is the Maximum A Posterior probability (MAP). Simulation analysis results show three conclusions: (1) The SNR of DSFH signal can be as low as –18 dB, which uses the matched SR detection. (2) Method for combining DSFH with the matched SR is suitable to detect the signals with SNR of –18 ～–14 dB. (3) In the case of –14 dB SNR, the DFSH signal detection performance increases by 25.47%, when using SR. The proposed method effectiveness is proved with simulation results.
Considering the problem that the Dual-Sequence Frequency Hopping (DSFH) can not communicate at extremely low Signal-to-Noise Ratio (SNR), a Stochastic Resonance (SR) detection method is proposed. The SR takes full advantage of the physical characteristics of DSFH signal to improve the detection performance. Firstly, the SR is constructed by analyzing signals of transmission, reception and the Intermediate Frequency (IF). The scale transaction is used to adjust the IF signal to fit the SR. Secondly, the non-autonomous Fokker-Plank Equation (FPE) is transformed into an autonomous equation by introducing the decision time. Therefore, the analytical solution of the probability density function with the parameter of decision time is obtained. Finally, the detection probability, false alarm probability and Receiver Operating Characteristics (ROC) curve are obtained, when the criterion is the Maximum A Posterior probability (MAP). Simulation analysis results show three conclusions: (1) The SNR of DSFH signal can be as low as –18 dB, which uses the matched SR detection. (2) Method for combining DSFH with the matched SR is suitable to detect the signals with SNR of –18 ～–14 dB. (3) In the case of –14 dB SNR, the DFSH signal detection performance increases by 25.47%, when using SR. The proposed method effectiveness is proved with simulation results.
2019, 41(10): 2350-2357.
doi: 10.11999/JEIT180832
Abstract:
To solve the problem that polarization sensitive array of defective electromagnetic vector sensor estimate multi parameter, a two-dimensional DOA estimation algorithm based on orthogonal dipole is proposed in this paper. First, eigendecomposition of the covariance matrix is produced by the received data vectors of the polarization sensitive array. Then the signal subspace is divided into four subarrays, and the phase difference between one of the subarray and the others is obtained according to the ESPRIT algorithm. Then the phase difference between different subarrays is paired. Finally, the DOA estimation and polarization parameters of the signal are calculated according to the phase difference. The uniform linear array composed by orthogonal dipoles can not be two-dimensional DOA estimated by using the MUSIC algorithm and the traditional ESPRIT algorithm. The algorithm proposed in this paper solves this problem, and compared with the polarization MUISC algorithm greatly reduces the complexity of the algorithm. The simulation results verify the effectiveness of the proposed algorithm.
To solve the problem that polarization sensitive array of defective electromagnetic vector sensor estimate multi parameter, a two-dimensional DOA estimation algorithm based on orthogonal dipole is proposed in this paper. First, eigendecomposition of the covariance matrix is produced by the received data vectors of the polarization sensitive array. Then the signal subspace is divided into four subarrays, and the phase difference between one of the subarray and the others is obtained according to the ESPRIT algorithm. Then the phase difference between different subarrays is paired. Finally, the DOA estimation and polarization parameters of the signal are calculated according to the phase difference. The uniform linear array composed by orthogonal dipoles can not be two-dimensional DOA estimated by using the MUSIC algorithm and the traditional ESPRIT algorithm. The algorithm proposed in this paper solves this problem, and compared with the polarization MUISC algorithm greatly reduces the complexity of the algorithm. The simulation results verify the effectiveness of the proposed algorithm.
2019, 41(10): 2358-2365.
doi: 10.11999/JEIT180858
Abstract:
Increasing the integration time can effectively improve the detection performance of passive radar. However, for maneuvering targets, the complex motions, such as high velocity, acceleration and jerk, cause existing detection methods to suffer the Range Migration (RM) and Doppler Frequency Migration (DFM) during the integration time, which deteriorates the detection performance. This paper addresses the long time coherent integration for a maneuvering target with high-order motion (e.g., jerk motion) in passive radar systems. A method based on Adjacent Cross Correlation Function (ACCF), Parameterized Centroid Frequency-Chirp Rate Distribution (PCFCRD) and Keystone Transform (KT)(ACCF-PCFCRD-KT), is proposed. Firstly, the signal model for the maneuvering targets is given, and the influence of the target velocity, acceleration and jerk on the coherent integration is analyzed. For the Doppler curvature induced by the jerk motion, the ACCF is firstly applied to reducing the order of RM and DFM. Then the PCFCRD operation is employed to estimate the acceleration and jerk parameters. After compensating the RM and DFM caused by the acceleration and jerk, the RM arising from the velocity is corrected via the KT operation and the target echo energy is coherently integrated. Simulation results demonstrate that the proposed method can effectively compensate the RM and DFM caused by the target motion parameters in passive radar, and for a maneuvering target with jerk motion, the proposed method achieves better integration performance over the existing methods.
Increasing the integration time can effectively improve the detection performance of passive radar. However, for maneuvering targets, the complex motions, such as high velocity, acceleration and jerk, cause existing detection methods to suffer the Range Migration (RM) and Doppler Frequency Migration (DFM) during the integration time, which deteriorates the detection performance. This paper addresses the long time coherent integration for a maneuvering target with high-order motion (e.g., jerk motion) in passive radar systems. A method based on Adjacent Cross Correlation Function (ACCF), Parameterized Centroid Frequency-Chirp Rate Distribution (PCFCRD) and Keystone Transform (KT)(ACCF-PCFCRD-KT), is proposed. Firstly, the signal model for the maneuvering targets is given, and the influence of the target velocity, acceleration and jerk on the coherent integration is analyzed. For the Doppler curvature induced by the jerk motion, the ACCF is firstly applied to reducing the order of RM and DFM. Then the PCFCRD operation is employed to estimate the acceleration and jerk parameters. After compensating the RM and DFM caused by the acceleration and jerk, the RM arising from the velocity is corrected via the KT operation and the target echo energy is coherently integrated. Simulation results demonstrate that the proposed method can effectively compensate the RM and DFM caused by the target motion parameters in passive radar, and for a maneuvering target with jerk motion, the proposed method achieves better integration performance over the existing methods.
2019, 41(10): 2366-2372.
doi: 10.11999/JEIT190014
Abstract:
Considering the disadvantage of oblique delay estimation of tropospheric scattering at arbitrary stations, which is difficult to obtain real-time sounding meteorological data, an oblique delay estimation algorithm of tropospheric scattering based on improved ray tracing method with ground meteorological parameters is proposed. In order to get rid of the method’s dependence on radiosonde data, the algorithm infers the relationship between refractive index and altitude through the formula of meteorological parameters in the model of medium latitude atmosphere. The interpolation of meteorological parameters in the model of UNB3m is used to gain the coefficient of temperature and water vapor pressure. Meteorological data for 2012 from 6 International GNSS Service (IGS) stations in Asia are selected to test the applicability of new method, the results suggest that precision is less than 1 cm. Then, the tropospheric slant delays of three parts observation stations under different angles of incidence (0°～5°) are calculated by the modified algorithm. The results suggest that the maximum delay is 17.03～33.10 m in a single way time transfer. In two way time transfer, when the delay can counteract 95%, time delay is 2.88～5.52 ns.
Considering the disadvantage of oblique delay estimation of tropospheric scattering at arbitrary stations, which is difficult to obtain real-time sounding meteorological data, an oblique delay estimation algorithm of tropospheric scattering based on improved ray tracing method with ground meteorological parameters is proposed. In order to get rid of the method’s dependence on radiosonde data, the algorithm infers the relationship between refractive index and altitude through the formula of meteorological parameters in the model of medium latitude atmosphere. The interpolation of meteorological parameters in the model of UNB3m is used to gain the coefficient of temperature and water vapor pressure. Meteorological data for 2012 from 6 International GNSS Service (IGS) stations in Asia are selected to test the applicability of new method, the results suggest that precision is less than 1 cm. Then, the tropospheric slant delays of three parts observation stations under different angles of incidence (0°～5°) are calculated by the modified algorithm. The results suggest that the maximum delay is 17.03～33.10 m in a single way time transfer. In two way time transfer, when the delay can counteract 95%, time delay is 2.88～5.52 ns.
2019, 41(10): 2373-2381.
doi: 10.11999/JEIT190098
Abstract:
Considering the problem that the scattering echo image of the new generation Doppler meteorological radar is reduced by the noise echoes such as non-rainfall, the accuracy of the refined short-term weather forecast is reduced. A method for semantic segmentation of meteorological radar noise image based on Deep Convolutional Neural Network(DCNN) is proposed. Firstly, a Deep Convolutional Neural Network Model (DCNNM) is designed. The training set data of the MJDATA data set are used for training, and the feature is extracted by the forward propagation process, and the high-dimensional global semantic information of the image is merged with the local feature details. Then, the network parameters are updated by using the training error value back propagation iteration to optimize the convergence effect of the model. Finally, the meteorological radar image data are segmented by the model. The experimental results show that the proposed method has better denoising effect on meteorological radar images, and compared with the optical flow method and the Fully Convolutional Networks (FCN), the method has high recognition accuracy for meteorological radar image real echo and noise echo, and the image pixel precision is high.
Considering the problem that the scattering echo image of the new generation Doppler meteorological radar is reduced by the noise echoes such as non-rainfall, the accuracy of the refined short-term weather forecast is reduced. A method for semantic segmentation of meteorological radar noise image based on Deep Convolutional Neural Network(DCNN) is proposed. Firstly, a Deep Convolutional Neural Network Model (DCNNM) is designed. The training set data of the MJDATA data set are used for training, and the feature is extracted by the forward propagation process, and the high-dimensional global semantic information of the image is merged with the local feature details. Then, the network parameters are updated by using the training error value back propagation iteration to optimize the convergence effect of the model. Finally, the meteorological radar image data are segmented by the model. The experimental results show that the proposed method has better denoising effect on meteorological radar images, and compared with the optical flow method and the Fully Convolutional Networks (FCN), the method has high recognition accuracy for meteorological radar image real echo and noise echo, and the image pixel precision is high.
2019, 41(10): 2382-2388.
doi: 10.11999/JEIT190095
Abstract:
The circuit structure optimization method for Basic programmable Logic Element (BLE) of FPGA is studied. Considering finding the solution to the bottleneck problem of low resource utilization efficiency in logic and arithmetic operations with 4-input Look Up Table (LUT), some efforts to improve BLE design based on 4-input LUT are explored. A high area-efficient LUT structure is proposed, and the possible benefits of such a new structure are analyzed theoretically and simulated. Further, a statistical method for evaluation of the post synthesis and mapping netlist is also proposed. Finally, a number of experiments are carried out to assess the proposed structure based on the MCNC and VTR benchmarks. The results show that, compared with Intel Stratix series FPGAs, the optimized structure proposed in this paper improves respectively the area efficiency of the FPGA by 10.428% and 10.433% in average under the MCNC and VTR benchmark circuits.
The circuit structure optimization method for Basic programmable Logic Element (BLE) of FPGA is studied. Considering finding the solution to the bottleneck problem of low resource utilization efficiency in logic and arithmetic operations with 4-input Look Up Table (LUT), some efforts to improve BLE design based on 4-input LUT are explored. A high area-efficient LUT structure is proposed, and the possible benefits of such a new structure are analyzed theoretically and simulated. Further, a statistical method for evaluation of the post synthesis and mapping netlist is also proposed. Finally, a number of experiments are carried out to assess the proposed structure based on the MCNC and VTR benchmarks. The results show that, compared with Intel Stratix series FPGAs, the optimized structure proposed in this paper improves respectively the area efficiency of the FPGA by 10.428% and 10.433% in average under the MCNC and VTR benchmark circuits.
2019, 41(10): 2389-2395.
doi: 10.11999/JEIT181134
Abstract:
To solve the problem of heat dissipation in Three Dimensional Field Programmable Gate Array Technology (3D FPGA), an interconnect channel architectural design method with low thermal gradient feature is proposed. A thermal resistance network model is established for the 3D FPGA, and theoretical studies and thermal simulation experiments are carried out on the influence of different types of channels on the thermal performance of 3D FPGA. Further, non-uniform vertical direction channel structures of 3D FPGA are proposed. Experiments indicate that 3D FPGA designed using the method proposed can reduce the maximum temperature gradient between different layers by 76.8% and the temperature gradient within the same layer by 10.4% compared with the traditional channel structure of 3D FPGA.
To solve the problem of heat dissipation in Three Dimensional Field Programmable Gate Array Technology (3D FPGA), an interconnect channel architectural design method with low thermal gradient feature is proposed. A thermal resistance network model is established for the 3D FPGA, and theoretical studies and thermal simulation experiments are carried out on the influence of different types of channels on the thermal performance of 3D FPGA. Further, non-uniform vertical direction channel structures of 3D FPGA are proposed. Experiments indicate that 3D FPGA designed using the method proposed can reduce the maximum temperature gradient between different layers by 76.8% and the temperature gradient within the same layer by 10.4% compared with the traditional channel structure of 3D FPGA.
2019, 41(10): 2396-2401.
doi: 10.11999/JEIT190051
Abstract:
Currently, the turbine air flow sensors are widely used to record the human exhalation signals in spirometry, but test results vary due to different expiratory flow for the same Forced Vital Capacity(FVC) measurements, and the differences are usually not in an acceptable range. To address this issue, a FVC velocity penalty model is proposed by introducing speed penalty items to the traditional mathematical model of turbine. Moreover, an over-amplitude drop sampling approach is used to calculate the rotations of the turbine due to the needs for the velocity penalty model to be able to accurately obtain the number of turbine rotations. The performance of the proposed approach is evaluated by using a syringe dispenser of 3L capacity, and results demonstrate that it can reduce the differences and meet the acceptable and accuracy criteria of the American Thoracic Society(ATS) and the European Respiratory Society(ERS) to some extent.
Currently, the turbine air flow sensors are widely used to record the human exhalation signals in spirometry, but test results vary due to different expiratory flow for the same Forced Vital Capacity(FVC) measurements, and the differences are usually not in an acceptable range. To address this issue, a FVC velocity penalty model is proposed by introducing speed penalty items to the traditional mathematical model of turbine. Moreover, an over-amplitude drop sampling approach is used to calculate the rotations of the turbine due to the needs for the velocity penalty model to be able to accurately obtain the number of turbine rotations. The performance of the proposed approach is evaluated by using a syringe dispenser of 3L capacity, and results demonstrate that it can reduce the differences and meet the acceptable and accuracy criteria of the American Thoracic Society(ATS) and the European Respiratory Society(ERS) to some extent.
2019, 41(10): 2402-2408.
doi: 10.11999/JEIT190003
Abstract:
Cipher cards play an important role in the field of information security. However, the performance of cipher cards are insufficient, and it is difficult to meet the needs of high-speed network security services. A design and system implementation method of high-speed PCIe cipher card based on MIPS64 multi-core processor is proposed, which supports the GM algorithm SM2/3/4 and international cryptographic algorithms, such as RSA, SHA and AES. The implemented system includes module of hardware, cryptographic algorithm, host driver and interface calling. An optimization scheme for the implementation of SM3 is proposed, the performance is improved by 19%. And the host to send requests in Non-Blocking mode is supported, so a single-process application can get the cipher card’s full load performance. Under 10-core CPU, the speed of SM2 signature and verification are 18000 and 4200 times/s, SM3 hash speed is 2200 Mbps, SM4 encryption/decryption speed is 8/10 Gbps, multiple indicators achieve higher level; When using 16-core CPU @1300 MHz, SM2/3 performance can be improved by more than 100%, and the speed of SM2 signature could achieve 105 times/s with 48-core CPU.
Cipher cards play an important role in the field of information security. However, the performance of cipher cards are insufficient, and it is difficult to meet the needs of high-speed network security services. A design and system implementation method of high-speed PCIe cipher card based on MIPS64 multi-core processor is proposed, which supports the GM algorithm SM2/3/4 and international cryptographic algorithms, such as RSA, SHA and AES. The implemented system includes module of hardware, cryptographic algorithm, host driver and interface calling. An optimization scheme for the implementation of SM3 is proposed, the performance is improved by 19%. And the host to send requests in Non-Blocking mode is supported, so a single-process application can get the cipher card’s full load performance. Under 10-core CPU, the speed of SM2 signature and verification are 18000 and 4200 times/s, SM3 hash speed is 2200 Mbps, SM4 encryption/decryption speed is 8/10 Gbps, multiple indicators achieve higher level; When using 16-core CPU @1300 MHz, SM2/3 performance can be improved by more than 100%, and the speed of SM2 signature could achieve 105 times/s with 48-core CPU.
2019, 41(10): 2409-2415.
doi: 10.11999/JEIT190038
Abstract:
In order to overcome the problem that cloud storage is not trusted and the low efficiency of ciphertext retrieval in cloud storage, a searchable ciphertext sorting encryption scheme based on B+ tree on the block chain is proposed. Combined with the blockchain technology, the problem of establishing reliable trust in multiple parties that do not understand each other is solved. A vector space model is used to reduce the complexity of the text and an efficient text retrieval system is implemented. The index structure of the B+ tree is used to improve the retrieval of ciphertext transactions on the blockchain. The ranking of multi-keyword query results is realized by the Term Frequency–Inverse Document Frequency (TF-IDF) algorithm. Under the random oracle model, it is proved that the scheme is adaptive and indistinguishable. Through the comparative analysis of efficiency, it is shown that the scheme achieves efficient ciphertext retrieval on the blockchain.
In order to overcome the problem that cloud storage is not trusted and the low efficiency of ciphertext retrieval in cloud storage, a searchable ciphertext sorting encryption scheme based on B+ tree on the block chain is proposed. Combined with the blockchain technology, the problem of establishing reliable trust in multiple parties that do not understand each other is solved. A vector space model is used to reduce the complexity of the text and an efficient text retrieval system is implemented. The index structure of the B+ tree is used to improve the retrieval of ciphertext transactions on the blockchain. The ranking of multi-keyword query results is realized by the Term Frequency–Inverse Document Frequency (TF-IDF) algorithm. Under the random oracle model, it is proved that the scheme is adaptive and indistinguishable. Through the comparative analysis of efficiency, it is shown that the scheme achieves efficient ciphertext retrieval on the blockchain.
2019, 41(10): 2416-2422.
doi: 10.11999/JEIT181113
Abstract:
With the development of the internet of things, small-scale communication devices such as wireless sensors and the Radio Frequency IDentification(RFID) tags are widely used, these micro-devices have limited computing power, so that the traditional cryptographic algorithms are difficult to implement on these devices. How to construct a high-efficiency diffusion layer becomes an urgent problem. With the best diffusion property, the Maximal Distance Separable (MDS) matrix is often used to construct the diffusion layer of block ciphers. The number of XOR operations (XORs) is an indicator of the efficiency of hardware applications. Combined with the XORs calculation method which can evaluate hardware efficiency more accurately and a matrix with special structure——Toeplitz matrix, efficient MDS matrices with less XORs can be constructed. Using the structural characteristics of the Toeplitz matrix, the constraints of matrix elements are improved, and the complexity of matrices searching is reduced. The 4×4 MDS matrices and the 6×6 MDS matrices with the least XORs in the finite field\begin{document}${\mathbb{F}_{{2^8}}}$\end{document} ![]()
![]()
With the development of the internet of things, small-scale communication devices such as wireless sensors and the Radio Frequency IDentification(RFID) tags are widely used, these micro-devices have limited computing power, so that the traditional cryptographic algorithms are difficult to implement on these devices. How to construct a high-efficiency diffusion layer becomes an urgent problem. With the best diffusion property, the Maximal Distance Separable (MDS) matrix is often used to construct the diffusion layer of block ciphers. The number of XOR operations (XORs) is an indicator of the efficiency of hardware applications. Combined with the XORs calculation method which can evaluate hardware efficiency more accurately and a matrix with special structure——Toeplitz matrix, efficient MDS matrices with less XORs can be constructed. Using the structural characteristics of the Toeplitz matrix, the constraints of matrix elements are improved, and the complexity of matrices searching is reduced. The 4×4 MDS matrices and the 6×6 MDS matrices with the least XORs in the finite field
2019, 41(10): 2423-2427.
doi: 10.11999/JEIT190070
Abstract:
GC weight is an important parameter of DNA code, and how to meet GC constant weight constraint DNA code is an interesting problem. In this paper, by establishing a bijection between DNA code and quaternion code, the DNA code that satisfies the GC constant weight constraint is converted into a GC constant weight quaternary code. Through the algebraic method, three types of DNA codes that meet the constant weight constraints of GC are constructed.
GC weight is an important parameter of DNA code, and how to meet GC constant weight constraint DNA code is an interesting problem. In this paper, by establishing a bijection between DNA code and quaternion code, the DNA code that satisfies the GC constant weight constraint is converted into a GC constant weight quaternary code. Through the algebraic method, three types of DNA codes that meet the constant weight constraints of GC are constructed.
2019, 41(10): 2428-2434.
doi: 10.11999/JEIT181160
Abstract:
Due to the feature of in-network caching in Named Data Networking (NDN), any consumer might fetch the cached contents from NDN routers, but the content producers have no idea about details of certain contents being accessed. Considering these problems, a fine-grained Traceable and Lightweight Access Control (TLAC) scheme is presented. In the TLAC scheme, an anonymous and secure " three-way handshake” authentication protocol is presented by collaboratively leveraging the combined public key and the Schnorr signature, and an improved secret sharing method is used to distribute the key efficiently. Finally, the experimental results prove the efficiency of TLAC scheme.
Due to the feature of in-network caching in Named Data Networking (NDN), any consumer might fetch the cached contents from NDN routers, but the content producers have no idea about details of certain contents being accessed. Considering these problems, a fine-grained Traceable and Lightweight Access Control (TLAC) scheme is presented. In the TLAC scheme, an anonymous and secure " three-way handshake” authentication protocol is presented by collaboratively leveraging the combined public key and the Schnorr signature, and an improved secret sharing method is used to distribute the key efficiently. Finally, the experimental results prove the efficiency of TLAC scheme.
2019, 41(10): 2435-2441.
doi: 10.11999/JEIT181130
Abstract:
Network Function Virtualization (NFV) brings flexibility and dynamics to the construction of service chain. However, the software and virtualization may cause security risks such as vulnerabilities and backdoors, which may have impact on Service Chain (SC) security. Thus, a Virtual Network Function (VNF) scheduling method is proposed. Firstly, heterogeneous images are built for every virtual network function in service chain, avoiding widespread attacks using common vulnerabilities. Then, one network function is selected dynamically and periodically. The executor of this network function is replaced by loading heterogeneous images. Finally, considering the impact of scheduling on the performance of network functions, Stackelberg game is used to model the attack and defense process, and the scheduling probability of each network function in the service chain is solved with the goal of optimizing the defender’s benefit. Experiments show that this method can reduce the rate of attacker’s success while controlling the overhead generated by the scheduling within an acceptable range.
Network Function Virtualization (NFV) brings flexibility and dynamics to the construction of service chain. However, the software and virtualization may cause security risks such as vulnerabilities and backdoors, which may have impact on Service Chain (SC) security. Thus, a Virtual Network Function (VNF) scheduling method is proposed. Firstly, heterogeneous images are built for every virtual network function in service chain, avoiding widespread attacks using common vulnerabilities. Then, one network function is selected dynamically and periodically. The executor of this network function is replaced by loading heterogeneous images. Finally, considering the impact of scheduling on the performance of network functions, Stackelberg game is used to model the attack and defense process, and the scheduling probability of each network function in the service chain is solved with the goal of optimizing the defender’s benefit. Experiments show that this method can reduce the rate of attacker’s success while controlling the overhead generated by the scheduling within an acceptable range.
2019, 41(10): 2442-2449.
doi: 10.11999/JEIT190093
Abstract:
Salient object detection which aims at automatically detecting what attracts human’s attention most in a scene, bootstrap learning based on Support Vector Machine(SVM) has achieved excellent performance in bottom-up methods. However, it is time-consuming for each image to be trained once based on multiple kernel SVM ensemble. So a salient object detection model via Weighted K-Nearest Neighbor Linear Blending (WKNNLB) is proposed. First of all, existing saliency detection methods are employed to generate weak saliency maps and obtain training samples. Then, Weighted K-Nearest Neighbor (WKNN) is introduced to learning salient score of samples. WKNN model needs no pre-training process, only needs selecting K value and computing saliency value by the K-nearest neighbors labels of training sample and the distances between the K-nearest neighbors training samples and the testing sample. In order to reduce the influence of selecting K value, linear blending of multi-WKNNs is applied to generating strong saliency maps. Finally, multi-scale saliency maps of weak and strong model are integrated together to further improve the detection performance. The experimental results on common ASD and complex DUT-OMRON datasets show that the algorithm is effective and superior in running time and performance. It can even perform favorable against the state-of-the-art methods when adopting better weak saliency map.
Salient object detection which aims at automatically detecting what attracts human’s attention most in a scene, bootstrap learning based on Support Vector Machine(SVM) has achieved excellent performance in bottom-up methods. However, it is time-consuming for each image to be trained once based on multiple kernel SVM ensemble. So a salient object detection model via Weighted K-Nearest Neighbor Linear Blending (WKNNLB) is proposed. First of all, existing saliency detection methods are employed to generate weak saliency maps and obtain training samples. Then, Weighted K-Nearest Neighbor (WKNN) is introduced to learning salient score of samples. WKNN model needs no pre-training process, only needs selecting K value and computing saliency value by the K-nearest neighbors labels of training sample and the distances between the K-nearest neighbors training samples and the testing sample. In order to reduce the influence of selecting K value, linear blending of multi-WKNNs is applied to generating strong saliency maps. Finally, multi-scale saliency maps of weak and strong model are integrated together to further improve the detection performance. The experimental results on common ASD and complex DUT-OMRON datasets show that the algorithm is effective and superior in running time and performance. It can even perform favorable against the state-of-the-art methods when adopting better weak saliency map.
2019, 41(10): 2450-2456.
doi: 10.11999/JEIT180857
Abstract:
A method of establishing a fingerprint database, which is based on distributed compressed sensing, is proposed to improve the low positioning accuracy and poor real-time positioning that exist in the current mine target positioning in China. Using the method, the fingerprint information of mine target fingerprint database can be reconstructed with high probability by collecting only a few fingerprint information (reference node IDs, Time Of Arrival (TOA) measurements based on electromagnetic wave and actual distance values) in the roadway in the off-line stage. Therefore, the data collection workload can be reduced and the work efficiency can be improved as well. In the subsequent on-line stage, according to the pattern matching method, the estimated distance between the target node and the reference nodes at the certain time can be obtained only by getting the reference node IDs and the real-time TOA measurements measured by the reference nodes at a certain moment, which guarantees the positioning accuracy and positioning real-time performance. Based on this method, an improved Compressive Sampling Modifying Matching Pursuit (CoSaMMP) algorithm is proposed to reconstruct the fingerprint information. The algorithm can effectively shorten the reconstruction time by using the folding method to increase the cutting force. The simulation results show that the proposed algorithm is feasible and effective.
A method of establishing a fingerprint database, which is based on distributed compressed sensing, is proposed to improve the low positioning accuracy and poor real-time positioning that exist in the current mine target positioning in China. Using the method, the fingerprint information of mine target fingerprint database can be reconstructed with high probability by collecting only a few fingerprint information (reference node IDs, Time Of Arrival (TOA) measurements based on electromagnetic wave and actual distance values) in the roadway in the off-line stage. Therefore, the data collection workload can be reduced and the work efficiency can be improved as well. In the subsequent on-line stage, according to the pattern matching method, the estimated distance between the target node and the reference nodes at the certain time can be obtained only by getting the reference node IDs and the real-time TOA measurements measured by the reference nodes at a certain moment, which guarantees the positioning accuracy and positioning real-time performance. Based on this method, an improved Compressive Sampling Modifying Matching Pursuit (CoSaMMP) algorithm is proposed to reconstruct the fingerprint information. The algorithm can effectively shorten the reconstruction time by using the folding method to increase the cutting force. The simulation results show that the proposed algorithm is feasible and effective.
2019, 41(10): 2457-2463.
doi: 10.11999/JEIT181121
Abstract:
The filtering performance of Gaussian Mixture Cardinality Balanced Multi-target Multi-Bernoulli (GM-CBMeMBer) filter can be effected by the heavy-tailed process noise and measurement noise. To solve this problem, a new STudent’s t Mixture Cardinality Balanced Multi-target Multi-Bernoulli (STM-CBMeMBer) filter is proposed. The process noise and measurement noise approximately obey the Student’s t distribution in the filter, where the Student’s t mixture model is used to describe approximately the posterior intensity of the multi-target. The predictive intensity and posterior intensity of Student’s t mixture form are deduced theoretically, and the closed recursive framework of cardinality balanced multi-target multi-Bernoulli filter is established. The simulation results show that, in the presence of the heavy-tailed process noise and the measurement noise, the filter can effectively suppress its interference, its tracking accuracy is superior over the traditional methods.
The filtering performance of Gaussian Mixture Cardinality Balanced Multi-target Multi-Bernoulli (GM-CBMeMBer) filter can be effected by the heavy-tailed process noise and measurement noise. To solve this problem, a new STudent’s t Mixture Cardinality Balanced Multi-target Multi-Bernoulli (STM-CBMeMBer) filter is proposed. The process noise and measurement noise approximately obey the Student’s t distribution in the filter, where the Student’s t mixture model is used to describe approximately the posterior intensity of the multi-target. The predictive intensity and posterior intensity of Student’s t mixture form are deduced theoretically, and the closed recursive framework of cardinality balanced multi-target multi-Bernoulli filter is established. The simulation results show that, in the presence of the heavy-tailed process noise and the measurement noise, the filter can effectively suppress its interference, its tracking accuracy is superior over the traditional methods.
2019, 41(10): 2464-2470.
doi: 10.11999/JEIT180971
Abstract:
To solve the problems of low robustness and tracking accuracy in target tracking when interference factors occur such as target fast motion and occlusion in complex video scenes, an Adaptive Strategy Fusion Target Tracking algorithm (ASFTT) is proposed based on multi-layer convolutional features. Firstly, the multi-layer convolutional features of frame images in Convolutional Neural Network(CNN) are extracted, which avoids the defect that the target information of the network is not comprehensive enough, so as to increase the generalization ability of the algorithm. Secondly, in order to improve the tracking accuracy of the algorithm, the multi-layer features are performed to calculate the correlation responses, which improves the tracking accuracy. Finally, the target position strategy in all responses are dynamically merged to locate the target through the adaptive strategy fusion algorithm in this paper. It comprehensively considers the historical strategy information and current strategy information of each responsive tracker to ensure the robustness. Experiments performed on the OTB2013 evaluation benchmark show that that the performance of the proposed algorithm are better than those of the other six state-of-the-art methods.
To solve the problems of low robustness and tracking accuracy in target tracking when interference factors occur such as target fast motion and occlusion in complex video scenes, an Adaptive Strategy Fusion Target Tracking algorithm (ASFTT) is proposed based on multi-layer convolutional features. Firstly, the multi-layer convolutional features of frame images in Convolutional Neural Network(CNN) are extracted, which avoids the defect that the target information of the network is not comprehensive enough, so as to increase the generalization ability of the algorithm. Secondly, in order to improve the tracking accuracy of the algorithm, the multi-layer features are performed to calculate the correlation responses, which improves the tracking accuracy. Finally, the target position strategy in all responses are dynamically merged to locate the target through the adaptive strategy fusion algorithm in this paper. It comprehensively considers the historical strategy information and current strategy information of each responsive tracker to ensure the robustness. Experiments performed on the OTB2013 evaluation benchmark show that that the performance of the proposed algorithm are better than those of the other six state-of-the-art methods.
2019, 41(10): 2471-2478.
doi: 10.11999/JEIT180867
Abstract:
To solve the problems of the existing in the process of human-computer interaction system, such as lack of emotion and low participation, a cognitive emotion interaction model based on game theory in PAD emotion space is proposed. Firstly, the interactive input emotion of participant is evaluated and some influence factors such as friendship and resonance are extracted to analyze the current human-computer interaction relationship. Secondly, modeling the emotional generation process of participants and robots by simulating the psychological game process in interpersonal communication, and the optimal emotional strategy of the robot is obtained by using the sub-game perfection equilibrium of the embedded game. Finally, the emotional state transition probability of the robot is updated according the optimal emotional strategy. The spatial coordinates of the six basic emotional states are used as labels to obtain the PAD spatial coordinate of the robot emotional state after emotional stimulate, The results of experiment show that compared with the others emotional interaction model, the proposed model can reduce the dependence of robots on external emotional stimuli and effective guide participants to participate in human-computer interaction, which provides some ideas for the emotion cognition model of robot in human-computer interaction.
To solve the problems of the existing in the process of human-computer interaction system, such as lack of emotion and low participation, a cognitive emotion interaction model based on game theory in PAD emotion space is proposed. Firstly, the interactive input emotion of participant is evaluated and some influence factors such as friendship and resonance are extracted to analyze the current human-computer interaction relationship. Secondly, modeling the emotional generation process of participants and robots by simulating the psychological game process in interpersonal communication, and the optimal emotional strategy of the robot is obtained by using the sub-game perfection equilibrium of the embedded game. Finally, the emotional state transition probability of the robot is updated according the optimal emotional strategy. The spatial coordinates of the six basic emotional states are used as labels to obtain the PAD spatial coordinate of the robot emotional state after emotional stimulate, The results of experiment show that compared with the others emotional interaction model, the proposed model can reduce the dependence of robots on external emotional stimuli and effective guide participants to participate in human-computer interaction, which provides some ideas for the emotion cognition model of robot in human-computer interaction.
2019, 41(10): 2479-2486.
doi: 10.11999/JEIT180963
Abstract:
In the case of high compression rates, the JPEG decompressed image can produce blocking artifacts, ringing effects and blurring, which affect seriously the visual effect of the image. In order to remove JPEG compression artifacts, a multi-scale dense residual network is proposed. Firstly, the proposed network introduces the dilate convolution into a dense block and uses different dilation factors to form multi-scale dense blocks. Then, the proposed network uses four multi-scale dense blocks to design the network into a structure with two branches, and the latter branch is used to supplement the features that are not extracted by the previous branch. Finally, the proposed network uses residual learning to improve network performance. In order to improve the versatility of the network, the network is trained by a joint training method with different compression quality factors, and a general model is trained for different compression quality factors. Experiments demonstrate that the proposed algorithm not only has high JPEG compression artifacts reduction performance, but also has strong generalization ability.
In the case of high compression rates, the JPEG decompressed image can produce blocking artifacts, ringing effects and blurring, which affect seriously the visual effect of the image. In order to remove JPEG compression artifacts, a multi-scale dense residual network is proposed. Firstly, the proposed network introduces the dilate convolution into a dense block and uses different dilation factors to form multi-scale dense blocks. Then, the proposed network uses four multi-scale dense blocks to design the network into a structure with two branches, and the latter branch is used to supplement the features that are not extracted by the previous branch. Finally, the proposed network uses residual learning to improve network performance. In order to improve the versatility of the network, the network is trained by a joint training method with different compression quality factors, and a general model is trained for different compression quality factors. Experiments demonstrate that the proposed algorithm not only has high JPEG compression artifacts reduction performance, but also has strong generalization ability.
2019, 41(10): 2487-2494.
doi: 10.11999/JEIT190017
Abstract:
High-resolution remote sensing images have complex visual contents, and extracting feature to represent image content accurately is the key to improving image retrieval performance. Convolutional Neural Networks (CNN) have strong transfer learning ability, and the high-level features of CNN can be efficiently transferred to high-resolution remote sensing images. In order to make full use of the advantages of high-level features, a combination and pooling method based on high-level feature maps is proposed to fuse high-level features from different CNNs. Firstly, the high-level features are adopted as special convolutional features to preserve the feature maps of the high-level outputs under different input sizes, and then the feature maps are combined into a larger feature map to integrate the features learned by different CNNs. The combined feature map is compressed by max-pooling method to extract salient features. Finally, the Principal Component Analysis (PCA) is utilized to reduce the redundancy of the salient features. The experimental results show that compared with the existing retrieval methods, the features extracted by this method have advantages in retrieval efficiency and precision.
High-resolution remote sensing images have complex visual contents, and extracting feature to represent image content accurately is the key to improving image retrieval performance. Convolutional Neural Networks (CNN) have strong transfer learning ability, and the high-level features of CNN can be efficiently transferred to high-resolution remote sensing images. In order to make full use of the advantages of high-level features, a combination and pooling method based on high-level feature maps is proposed to fuse high-level features from different CNNs. Firstly, the high-level features are adopted as special convolutional features to preserve the feature maps of the high-level outputs under different input sizes, and then the feature maps are combined into a larger feature map to integrate the features learned by different CNNs. The combined feature map is compressed by max-pooling method to extract salient features. Finally, the Principal Component Analysis (PCA) is utilized to reduce the redundancy of the salient features. The experimental results show that compared with the existing retrieval methods, the features extracted by this method have advantages in retrieval efficiency and precision.
2019, 41(10): 2495-2500.
doi: 10.11999/JEIT181195
Abstract:
A lossy frame memory compression algorithm using Direction Interpolation Prediction Variable Length Coding (DIPVLC) is proposed to improve frame memory compression performance. Firstly, the prediction residual is obtained by adaptive texture directional interpolation. Then, a new rate-distortion is optimized to quantize prediction residual. Finally, the run length Golomb method is used to entropy coding for quantized residual. Simulation results show that compared with parallel Content Aware Adaptive Quantization (CAAQ) oriented lossy frame memory recompression for HEVC, the proposed algorithm improves the compression rate by 10.05% and reduces the encoding time by 10.62% with less PSNR reduction.
A lossy frame memory compression algorithm using Direction Interpolation Prediction Variable Length Coding (DIPVLC) is proposed to improve frame memory compression performance. Firstly, the prediction residual is obtained by adaptive texture directional interpolation. Then, a new rate-distortion is optimized to quantize prediction residual. Finally, the run length Golomb method is used to entropy coding for quantized residual. Simulation results show that compared with parallel Content Aware Adaptive Quantization (CAAQ) oriented lossy frame memory recompression for HEVC, the proposed algorithm improves the compression rate by 10.05% and reduces the encoding time by 10.62% with less PSNR reduction.
2019, 41(10): 2501-2508.
doi: 10.11999/JEIT190036
Abstract:
The Fast Super-Resolution Convolutional Neural Network algorithm (FSRCNN) is difficult to extract deep image information due to the small number of convolution layers and the correlation lack between the feature information of adjacent convolutional layers. To solve this problem, a deep residual network super-resolution reconstruction method with multi-level skip connections is proposed. Firstly, a residual block with multi-level skip connections is designed to solve the problem that the characteristic information of adjacent convolutional layers lacks relevance. A deep residual network with multi-level skip connections is constructed on the basis of the residual block. Then, the deep residual network connected to the multi-level skip is trained by using the adaptive gradient rate strategy of Stochastic Gradient Descent (SGD) method and the network super-resolution reconstruction model is obtained. Finally, the low-resolution image is input into the deep residual network super-resolution reconstruction model with the multi-level skip connections, and the residual eigenvalue is obtained by the residual block connected the multi-level skip connections. The residual eigenvalue and the low resolution image are combined and converted into a high resolution image. The proposed method is compared with the bicubic, A+, SRCNN, FSRCNN and ESPCN algorithms in the Set5 and Set14 test sets. The proposed method is superior to other comparison algorithms in terms of visual effects and evaluation index values.
The Fast Super-Resolution Convolutional Neural Network algorithm (FSRCNN) is difficult to extract deep image information due to the small number of convolution layers and the correlation lack between the feature information of adjacent convolutional layers. To solve this problem, a deep residual network super-resolution reconstruction method with multi-level skip connections is proposed. Firstly, a residual block with multi-level skip connections is designed to solve the problem that the characteristic information of adjacent convolutional layers lacks relevance. A deep residual network with multi-level skip connections is constructed on the basis of the residual block. Then, the deep residual network connected to the multi-level skip is trained by using the adaptive gradient rate strategy of Stochastic Gradient Descent (SGD) method and the network super-resolution reconstruction model is obtained. Finally, the low-resolution image is input into the deep residual network super-resolution reconstruction model with the multi-level skip connections, and the residual eigenvalue is obtained by the residual block connected the multi-level skip connections. The residual eigenvalue and the low resolution image are combined and converted into a high resolution image. The proposed method is compared with the bicubic, A+, SRCNN, FSRCNN and ESPCN algorithms in the Set5 and Set14 test sets. The proposed method is superior to other comparison algorithms in terms of visual effects and evaluation index values.
2019, 41(10): 2509-2515.
doi: 10.11999/JEIT180957
Abstract:
Considering the problem that the prediction accuracy is not accurate enough when the depth information is recovered from the monocular vision image, a method of depth estimation of road scenes based on pyramid pooling network is proposed. Firstly, using a combination of four residual network blocks, the road scene image features are extracted, and then through the sampling, the features are gradually restored to the original image size, and the depth of the residual block is increased. Considering the diversity of information in different scales, the features with same sizes extracted from the sampling process and the feature extraction process are merged. In addition, pyramid pooling network blocks are added to the advanced features extracted by four residual network blocks for scene analysis, and the feature graph output of pyramid pooling network blocks is finally restored to the original image size and input prediction layer together with the output of the upper sampling module. Through experiments on KITTI data set, the results show that the proposed method is superior to the existing method.
Considering the problem that the prediction accuracy is not accurate enough when the depth information is recovered from the monocular vision image, a method of depth estimation of road scenes based on pyramid pooling network is proposed. Firstly, using a combination of four residual network blocks, the road scene image features are extracted, and then through the sampling, the features are gradually restored to the original image size, and the depth of the residual block is increased. Considering the diversity of information in different scales, the features with same sizes extracted from the sampling process and the feature extraction process are merged. In addition, pyramid pooling network blocks are added to the advanced features extracted by four residual network blocks for scene analysis, and the feature graph output of pyramid pooling network blocks is finally restored to the original image size and input prediction layer together with the output of the upper sampling module. Through experiments on KITTI data set, the results show that the proposed method is superior to the existing method.
2019, 41(10): 2516-2524.
doi: 10.11999/JEIT181059
Abstract:
In partial label learning, the true label of an instance is hidden in a label-set consisting of a group of candidate labels. The existing partial label learning algorithm only measures the similarity between instances based on feature vectors and lacks the utilization of the candidate labelset information. In this paper, a Candidate Label-Aware Partial Label Learning (CLAPLL) method is proposed, which combines effectively candidate label information to measure the similarity between instances during the graph construction phase. First, based on the jaccard distance and linear reconstruction, the similarity between the candidate labelsets of instances is calculated. Then, the similarity graph is constructed by combining the similarity of the instances and the label-sets, and then the existing graph-based partial label learning algorithm is presented for learning and prediction. The experimental results on 3 synthetic datasets and 6 real datasets show that disambiguation accuracy of the proposed method is 0.3%～16.5% higher than baseline algorithm, and the classification accuracy is increased by 0.2%～2.8%.
In partial label learning, the true label of an instance is hidden in a label-set consisting of a group of candidate labels. The existing partial label learning algorithm only measures the similarity between instances based on feature vectors and lacks the utilization of the candidate labelset information. In this paper, a Candidate Label-Aware Partial Label Learning (CLAPLL) method is proposed, which combines effectively candidate label information to measure the similarity between instances during the graph construction phase. First, based on the jaccard distance and linear reconstruction, the similarity between the candidate labelsets of instances is calculated. Then, the similarity graph is constructed by combining the similarity of the instances and the label-sets, and then the existing graph-based partial label learning algorithm is presented for learning and prediction. The experimental results on 3 synthetic datasets and 6 real datasets show that disambiguation accuracy of the proposed method is 0.3%～16.5% higher than baseline algorithm, and the classification accuracy is increased by 0.2%～2.8%.
2019, 41(10): 2525-2531.
doi: 10.11999/JEIT190047
Abstract:
Utilizing multiple data (elevation information) to assist remote sensing image segmentation is an important research topic in recent years. However, the existing methods usually directly use multivariate data as the input of the model, which fails to make full use of the multi-level features. In addition, the target size varies in remote sensing images, for some small targets, such as vehicles, houses, etc., it is difficult to achieve detailed segmentation. Considering these problems, a Multi-Feature map Pyramid fusion deep Network (MFPNet) is proposed, which utilizes optical remote sensing images and elevation data as input to extract multi-level features from images. Then the pyramid pooling structure is introduced to extract the multi-scale features from different levels. Finally, a multi-level and multi-scale feature fusion strategy is designed, which utilizes comprehensively the feature information of multivariate data to achieve detailed segmentation of remote sensing images. Experiment results on the Vaihingen dataset demonstrate the effectiveness of the proposed method.
Utilizing multiple data (elevation information) to assist remote sensing image segmentation is an important research topic in recent years. However, the existing methods usually directly use multivariate data as the input of the model, which fails to make full use of the multi-level features. In addition, the target size varies in remote sensing images, for some small targets, such as vehicles, houses, etc., it is difficult to achieve detailed segmentation. Considering these problems, a Multi-Feature map Pyramid fusion deep Network (MFPNet) is proposed, which utilizes optical remote sensing images and elevation data as input to extract multi-level features from images. Then the pyramid pooling structure is introduced to extract the multi-scale features from different levels. Finally, a multi-level and multi-scale feature fusion strategy is designed, which utilizes comprehensively the feature information of multivariate data to achieve detailed segmentation of remote sensing images. Experiment results on the Vaihingen dataset demonstrate the effectiveness of the proposed method.
2019, 41(10): 2532-2540.
doi: 10.11999/JEIT190101
Abstract:
Considering the inaccurate description of feature differences between nodes in the graph-based saliency detection algorithm, an image saliency detection algorithm combining object compactness and regional homogeneity strategy is proposed. Different from the commonly used graph-based model, a sparse graph-based structure closer to the human visual system and a novel regional homogeneity graph-based structure are established. They are used to describe the correlation within the foreground and the difference between foreground and background. Therefore, many redundant connections of nodes are eliminated and the local spatial relationship of nodes is strengthened. Then the clusters are combined to form a saliency map by means of manifold ranking. Finally, the background confidence is introduced for saliency optimization by the similarity of the background region clusters and the final detection result is obtained. Compared with 4 popular graph-based algorithms on the four benchmark datasets, the proposed algorithm can highlight the salient regions clearly and has better performance in the evaluation of multiple comprehensive indicators.
Considering the inaccurate description of feature differences between nodes in the graph-based saliency detection algorithm, an image saliency detection algorithm combining object compactness and regional homogeneity strategy is proposed. Different from the commonly used graph-based model, a sparse graph-based structure closer to the human visual system and a novel regional homogeneity graph-based structure are established. They are used to describe the correlation within the foreground and the difference between foreground and background. Therefore, many redundant connections of nodes are eliminated and the local spatial relationship of nodes is strengthened. Then the clusters are combined to form a saliency map by means of manifold ranking. Finally, the background confidence is introduced for saliency optimization by the similarity of the background region clusters and the final detection result is obtained. Compared with 4 popular graph-based algorithms on the four benchmark datasets, the proposed algorithm can highlight the salient regions clearly and has better performance in the evaluation of multiple comprehensive indicators.
