Email alert
2019 Vol. 41, No. 6
column
- contents
- Radar Signal Processing
- Electromagnetic Field and Electromagnetic Wave Technology
- Wireless Communication and Internet of Things
- Network and Information Security
- Circuit and System Design
- Pattern Recognition and Intelligent Information Processing
- Information of National Natural Science Foundation of China
Display Method:
2019, 41(6): 1271-1279.
doi: 10.11999/JEIT180661
Abstract:
Microwave photonics radar generates signals with large bandwidth and small wavelength. It has capability of ultra-high resolution of Inverse Synthetic Aperture Radar(ISAR) image. Because the approximation of rotational components is not tenable, traditional ISAR imaging algorithm is not suitable to microwave photonics radar. In the microwave photonics radar imaging, the rotational components result in range curvature and quadratic phase error changing with distance. To solve this problem, an effective ISAR imaging algorithm is put forward which considers the influence of the target’s rotational component to echo envelope and phase. The value of envelope correlation is take as objective function and the target’s rotate speed is estimated by iteration; The range curvature is corrected by time resampling; The quadratic phase error is compensated by azimuth compensation function. Both simulated and real-measured data experimental results confirm the effectiveness of the proposed algorithm.
Microwave photonics radar generates signals with large bandwidth and small wavelength. It has capability of ultra-high resolution of Inverse Synthetic Aperture Radar(ISAR) image. Because the approximation of rotational components is not tenable, traditional ISAR imaging algorithm is not suitable to microwave photonics radar. In the microwave photonics radar imaging, the rotational components result in range curvature and quadratic phase error changing with distance. To solve this problem, an effective ISAR imaging algorithm is put forward which considers the influence of the target’s rotational component to echo envelope and phase. The value of envelope correlation is take as objective function and the target’s rotate speed is estimated by iteration; The range curvature is corrected by time resampling; The quadratic phase error is compensated by azimuth compensation function. Both simulated and real-measured data experimental results confirm the effectiveness of the proposed algorithm.
2019, 41(6): 1280-1286.
doi: 10.11999/JEIT181086
Abstract:
The equivalent rotation center should be estimated accurately in the Inverse Synthetic Aperture Radar (ISAR) for the issue of image defocusing induced by the Migration Through Resolution Cells (MTRC). In this paper, an equivalent rotation center estimation algorithm based on image rotation and correlation is proposed for the space target. First, the instantaneous imaging mechanism of ISAR is analyzed. Second, two images with different observation angles are obtained by using the echo data with the same motion compensation algorithm. Finally, the equivalent rotation center is estimated based on the scaled image pixel rotation and image correlation. Consequently, the estimated position of the rotation center is obtained, when the assumed rotation center is in accordance with the real one and the maximum correlation coefficient of two images is achieved. The results demonstrate the effectiveness and robustness of the proposed algorithm.
The equivalent rotation center should be estimated accurately in the Inverse Synthetic Aperture Radar (ISAR) for the issue of image defocusing induced by the Migration Through Resolution Cells (MTRC). In this paper, an equivalent rotation center estimation algorithm based on image rotation and correlation is proposed for the space target. First, the instantaneous imaging mechanism of ISAR is analyzed. Second, two images with different observation angles are obtained by using the echo data with the same motion compensation algorithm. Finally, the equivalent rotation center is estimated based on the scaled image pixel rotation and image correlation. Consequently, the estimated position of the rotation center is obtained, when the assumed rotation center is in accordance with the real one and the maximum correlation coefficient of two images is achieved. The results demonstrate the effectiveness and robustness of the proposed algorithm.
2019, 41(6): 1287-1293.
doi: 10.11999/JEIT180663
Abstract:
The moving target component is often defocused in spaceborne SAR images. Therefore, the moving target detection performance is affected depending on the degree of defocusing. Combined with the RD algorithm, a moving-targets detection algorithm for spaceborne SAR based on a two-dimensional velocity search is proposed. Through velocity search on the distance direction and the azimuth direction, the Doppler parameters of possible moving targets can be matched. Then the strongest value among all the searching velocity results for each pixel is used for Constant False Alarm Rate(CFAR) detector. This core process can improve the detection performance of moving target component. Simulation results validate the effectiveness of the proposed method.
The moving target component is often defocused in spaceborne SAR images. Therefore, the moving target detection performance is affected depending on the degree of defocusing. Combined with the RD algorithm, a moving-targets detection algorithm for spaceborne SAR based on a two-dimensional velocity search is proposed. Through velocity search on the distance direction and the azimuth direction, the Doppler parameters of possible moving targets can be matched. Then the strongest value among all the searching velocity results for each pixel is used for Constant False Alarm Rate(CFAR) detector. This core process can improve the detection performance of moving target component. Simulation results validate the effectiveness of the proposed method.
2019, 41(6): 1294-1301.
doi: 10.11999/JEIT180145
Abstract:
In Compressive Sensing (CS) imaging algorithms, the true targets usually can not locate on the pre-defined grids exactly. Such Off-grid problems result in mismatch between true echo and measurement matrix, which seriously degrades the performance of radar imaging. An adaptive calibration method is proposed to solve the off-grid problems in MIMO radar Three-Dimensional (3D) imaging. Bayesian probability density functions can be constructed based on the sparse echo model of Off-grid targets, and the Maximum A Posteriori (MAP) method is used to obtain sparse imaging with mismatch errors. Compared with the traditional methods, the proposed method can make full use of mismatch parameters’ priori information and adaptively update the parameters, which can reduce the influence of mismatch errors, and achieve high-precision estimation for sparse targets and noise power. Finally, the simulation results confirm that the proposed method can effectively optimize mismatch errors with accurate and stable imaging performance.
In Compressive Sensing (CS) imaging algorithms, the true targets usually can not locate on the pre-defined grids exactly. Such Off-grid problems result in mismatch between true echo and measurement matrix, which seriously degrades the performance of radar imaging. An adaptive calibration method is proposed to solve the off-grid problems in MIMO radar Three-Dimensional (3D) imaging. Bayesian probability density functions can be constructed based on the sparse echo model of Off-grid targets, and the Maximum A Posteriori (MAP) method is used to obtain sparse imaging with mismatch errors. Compared with the traditional methods, the proposed method can make full use of mismatch parameters’ priori information and adaptively update the parameters, which can reduce the influence of mismatch errors, and achieve high-precision estimation for sparse targets and noise power. Finally, the simulation results confirm that the proposed method can effectively optimize mismatch errors with accurate and stable imaging performance.
2019, 41(6): 1302-1309.
doi: 10.11999/JEIT180677
Abstract:
In order to meet the demand for high real-time and high generalization performance of radar recognition, a radar High Resolution Range Profile (HRRP) recognition method based on deep multi-scale one dimension convolutional neural network is proposed. The multi-scale convolutional layer that can represent the complex features of HRRP is designed based on two features of the convolution kernels which are weight sharing and extraction of different fineness features from different scales, respectively. At last, the center loss function is used to improve the separability of features. Experimental results show that the model can greatly improve the accuracy of the target recognition under non-ideal conditions and solve the problem of the target aspect sensitivity, which also has good robustness and generalization performance.
In order to meet the demand for high real-time and high generalization performance of radar recognition, a radar High Resolution Range Profile (HRRP) recognition method based on deep multi-scale one dimension convolutional neural network is proposed. The multi-scale convolutional layer that can represent the complex features of HRRP is designed based on two features of the convolution kernels which are weight sharing and extraction of different fineness features from different scales, respectively. At last, the center loss function is used to improve the separability of features. Experimental results show that the model can greatly improve the accuracy of the target recognition under non-ideal conditions and solve the problem of the target aspect sensitivity, which also has good robustness and generalization performance.
2019, 41(6): 1310-1316.
doi: 10.11999/JEIT180714
Abstract:
To address track-to-track association problem of radar and Electronic Support Measurements (ESM) in the presence of sensor biases and different targets reported by different sensors, an anti-bias track-to-track association algorithm based on track vectors hierarchical clustering is proposed. Firstly, the equivalent measurement is derived in the Modified Polar Coordinates (MPC). Linear relationship between state estimates and real states, sensor biases, measurement errors are established based on the approximate expansion of the equivalent measurement. The track vectors are obtained by the real state cancellation method. The homologous tracks are extracted by the method of track vectors hierarchical clustering, according to the statistical characteristics of Gaussian random vectors. The effectiveness of the proposed algorithm is verified by Monte Carlo simulation experiments in the presence of sensor biases, targets densities and detection probabilities.
To address track-to-track association problem of radar and Electronic Support Measurements (ESM) in the presence of sensor biases and different targets reported by different sensors, an anti-bias track-to-track association algorithm based on track vectors hierarchical clustering is proposed. Firstly, the equivalent measurement is derived in the Modified Polar Coordinates (MPC). Linear relationship between state estimates and real states, sensor biases, measurement errors are established based on the approximate expansion of the equivalent measurement. The track vectors are obtained by the real state cancellation method. The homologous tracks are extracted by the method of track vectors hierarchical clustering, according to the statistical characteristics of Gaussian random vectors. The effectiveness of the proposed algorithm is verified by Monte Carlo simulation experiments in the presence of sensor biases, targets densities and detection probabilities.
2019, 41(6): 1317-1323.
doi: 10.11999/JEIT180655
Abstract:
The system biases degrade seriously the location precision for the multi-static passive radar system. A joint registration and passive localization algorithm based on Constrained Total Least Squares (CTLS) using Direction Of Arrival (DOA) and Time Difference Of Arrival (TDOA) measurements is developed to address the multi-static radar localization problem under the influence of system biases. Firstly, the nonlinear DOA and TDOA measurement equations are linearized by introducing auxiliary variables. Considering the statistical correlation properties of the noise matrix in the pseudo-linear equations, a joint biases registration and passive localization problem is formulated as a CTLS problem and the Newton’s method is applied to solving the CTLS problem. Moreover, a dependent least squares algorithm is designed to improve the target position estimation using the relationship between auxiliary variables and target position. An iterative post-estimate procedure is exploited to enhance further the estimation accuracy of the system biases. Finally, the theoretical error of the proposed algorithm is derived. Simulations demonstrate that the proposed algorithm can effectively estimate the system biases and target position.
The system biases degrade seriously the location precision for the multi-static passive radar system. A joint registration and passive localization algorithm based on Constrained Total Least Squares (CTLS) using Direction Of Arrival (DOA) and Time Difference Of Arrival (TDOA) measurements is developed to address the multi-static radar localization problem under the influence of system biases. Firstly, the nonlinear DOA and TDOA measurement equations are linearized by introducing auxiliary variables. Considering the statistical correlation properties of the noise matrix in the pseudo-linear equations, a joint biases registration and passive localization problem is formulated as a CTLS problem and the Newton’s method is applied to solving the CTLS problem. Moreover, a dependent least squares algorithm is designed to improve the target position estimation using the relationship between auxiliary variables and target position. An iterative post-estimate procedure is exploited to enhance further the estimation accuracy of the system biases. Finally, the theoretical error of the proposed algorithm is derived. Simulations demonstrate that the proposed algorithm can effectively estimate the system biases and target position.
2019, 41(6): 1324-1329.
doi: 10.11999/JEIT180672
Abstract:
As for super resolution Direction-of-Arrival (DoA) estimation with random arrays, it is still challenged to obtain efficient and statistically unbiased estimates. Based on Augmented Estimation of Signal Parameters via Rotational Invariance Techniques (AESPRIT), an efficient DoA estimation algorithm is proposed for random arrays. AESPRIT is modified with a closed loop structure; Array Interpolation Technique (AIT) is utilized to provide the initial phase compensation angles to the loop. Therefore, the final DoA estimates can be calculated efficiently and accurately through iteration. The proposed algorithm gives algebraic solution directly and has low computational complexity. At the same time, the results are statistically unbiased because no manifold mapping or mode truncation error is introduced. Simulations verify the effectiveness of the proposed algorithm.
As for super resolution Direction-of-Arrival (DoA) estimation with random arrays, it is still challenged to obtain efficient and statistically unbiased estimates. Based on Augmented Estimation of Signal Parameters via Rotational Invariance Techniques (AESPRIT), an efficient DoA estimation algorithm is proposed for random arrays. AESPRIT is modified with a closed loop structure; Array Interpolation Technique (AIT) is utilized to provide the initial phase compensation angles to the loop. Therefore, the final DoA estimates can be calculated efficiently and accurately through iteration. The proposed algorithm gives algebraic solution directly and has low computational complexity. At the same time, the results are statistically unbiased because no manifold mapping or mode truncation error is introduced. Simulations verify the effectiveness of the proposed algorithm.
2019, 41(6): 1330-1335.
doi: 10.11999/JEIT180648
Abstract:
A metameterial absorber is designed, fabricated and experimentally demonstrated to realized ultra-wideband absorption based on loading lumped resistances to raise the efficiency of absorber. The proposed structure comprises of an upper absorber and an under absorber by longitudinal cascade to expand bandwidth. The analysis of equivalent circuit show that the absorber has good impedance matching in a wide frequency band and the mechanism of wave absorption is verified by current analysis. The size of the unit is only about 0.089\begin{document}$\lambda_ {\rm{L}}$\end{document} ![]()
![]()
\begin{document}$\lambda_ {\rm{L}}$\end{document} ![]()
![]()
\begin{document}$\lambda_ {\rm{L}}$\end{document} ![]()
![]()
\begin{document}$\lambda_ {\rm{L}}$\end{document} ![]()
![]()
A metameterial absorber is designed, fabricated and experimentally demonstrated to realized ultra-wideband absorption based on loading lumped resistances to raise the efficiency of absorber. The proposed structure comprises of an upper absorber and an under absorber by longitudinal cascade to expand bandwidth. The analysis of equivalent circuit show that the absorber has good impedance matching in a wide frequency band and the mechanism of wave absorption is verified by current analysis. The size of the unit is only about 0.089
2019, 41(6): 1336-1343.
doi: 10.11999/JEIT180247
Abstract:
Transformation Optics (TO) is a hot topic in the research area of electrical-magnetic fields. For providing further theoretical support to the design of stealth carpet based on TO, three basic mathematic problems of TO are discussed in this paper. Firstly, the uniqueness of transformation form in three-dimensional transformation of Maxwell’s equations is analyzed. A new transformation model is proposed, which is different from the classical one shown in reference. The new model also leads to a new transformation method that can generate flexible characteristic impedance in transformation space. Based on this, a design method of stealth cloak or carpet that can be used to hide the target in an area surrounded by medium with given permittivity is discussed. During this process, only the field distribution in free space is required as the original field during mapping. Secondly, the two-dimensional transformation of the wave equation is studied. The transformation of the magnetic field component in the two-dimensional transformation based on the wave equation of the electric field component is analyzed. The boundary matching during transformation is also discussed. The two dimension design method of stealth cloak or carpet that can be used to hide a target in an area surrounded by medium with specified permittivity is also discussed. Finally, the sufficiency and necessity of conformal transformation for designing a two dimension stealth cloak with non-uniform and anisotropic medium are proved strictly. The simulation results of a stealth carpet embedded in material are given to verify the proposed method. The analysis and the related conclusion presented in the paper provide theoretical support to the related application based on TO.
Transformation Optics (TO) is a hot topic in the research area of electrical-magnetic fields. For providing further theoretical support to the design of stealth carpet based on TO, three basic mathematic problems of TO are discussed in this paper. Firstly, the uniqueness of transformation form in three-dimensional transformation of Maxwell’s equations is analyzed. A new transformation model is proposed, which is different from the classical one shown in reference. The new model also leads to a new transformation method that can generate flexible characteristic impedance in transformation space. Based on this, a design method of stealth cloak or carpet that can be used to hide the target in an area surrounded by medium with given permittivity is discussed. During this process, only the field distribution in free space is required as the original field during mapping. Secondly, the two-dimensional transformation of the wave equation is studied. The transformation of the magnetic field component in the two-dimensional transformation based on the wave equation of the electric field component is analyzed. The boundary matching during transformation is also discussed. The two dimension design method of stealth cloak or carpet that can be used to hide a target in an area surrounded by medium with specified permittivity is also discussed. Finally, the sufficiency and necessity of conformal transformation for designing a two dimension stealth cloak with non-uniform and anisotropic medium are proved strictly. The simulation results of a stealth carpet embedded in material are given to verify the proposed method. The analysis and the related conclusion presented in the paper provide theoretical support to the related application based on TO.
2019, 41(6): 1344-1351.
doi: 10.11999/JEIT180680
Abstract:
Cooperative MIMO technology can transform interference signals into useful signals by means of cooperative transmission or reception. It can solve the echo channel effect and improve the system capacity to be introduced into high-speed railway wireless communication. To master the channel characteristics of cooperative MIMO technology in high-speed railway scenarios, based on the geometric stochastic scattering theories, a new channel model for cooperative MIMO channel in high-speed railway scenarios is proposed, which can be applied to multiple high-speed railway scenarios by simply adjusting its several key parameters. Based on this model, the channel impulse response is calculated, the multi-link spatial correlation function is derived, the numerical calculation, simulation analysis and verification of measured data are carried out. Simulation results show that the multi-link spatial correlation is stronger when the LOS component is stronger and the angle spread of scattered components is smaller. The components which are scattered less times have a stronger spatial correlation. The theoretical model is verified by the measured data of the LTE special network of the Beijing-Tianjin high-speed railway section. These conclusions contribute to understanding the cooperative MIMO channels and conducting effective measurement activities.
Cooperative MIMO technology can transform interference signals into useful signals by means of cooperative transmission or reception. It can solve the echo channel effect and improve the system capacity to be introduced into high-speed railway wireless communication. To master the channel characteristics of cooperative MIMO technology in high-speed railway scenarios, based on the geometric stochastic scattering theories, a new channel model for cooperative MIMO channel in high-speed railway scenarios is proposed, which can be applied to multiple high-speed railway scenarios by simply adjusting its several key parameters. Based on this model, the channel impulse response is calculated, the multi-link spatial correlation function is derived, the numerical calculation, simulation analysis and verification of measured data are carried out. Simulation results show that the multi-link spatial correlation is stronger when the LOS component is stronger and the angle spread of scattered components is smaller. The components which are scattered less times have a stronger spatial correlation. The theoretical model is verified by the measured data of the LTE special network of the Beijing-Tianjin high-speed railway section. These conclusions contribute to understanding the cooperative MIMO channels and conducting effective measurement activities.
2019, 41(6): 1352-1358.
doi: 10.11999/JEIT180228
Abstract:
A massive MIMO full-duplex relaying system is considerd in this paper, in which multiple single-antenna sources simultaneously communicate with multiple single-antenna destinations using a single relay that is equipped with\begin{document}${N_{{\mathop{\rm rx}\nolimits} }}$\end{document} ![]()
![]()
\begin{document}${N_{{\mathop{\rm tx}\nolimits} }}$\end{document} ![]()
![]()
A massive MIMO full-duplex relaying system is considerd in this paper, in which multiple single-antenna sources simultaneously communicate with multiple single-antenna destinations using a single relay that is equipped with
2019, 41(6): 1359-1364.
doi: 10.11999/JEIT180762
Abstract:
Simultaneous Wireless Information and Power Transfer (SWIPT) is an effective technique to solve the energy limitation problem of wireless networks. A multi-carrier SWIPT communication system that includes one Base Station (BS) and multiple users is investigated. Both the uplink and downlink of the system apply the OFDM transmission. In the downlink, the BS transmits information and power over different subcarriers to the users simultaneously. In the uplink, the user transmits information to the BS by using the power harvested from the BS in the downlink. This paper aims to maximize the weighted sum of the downlink and uplink achievable rates by jointly optimizing subcarrier allocation and power allocation of the uplink and downlink. An optimal algorithm is proposed to solve this resulted optimization problem, which is based on the Lagrange duality method and the ellipsoid method. The performances of the proposed algorithm are verified by computer simulations.
Simultaneous Wireless Information and Power Transfer (SWIPT) is an effective technique to solve the energy limitation problem of wireless networks. A multi-carrier SWIPT communication system that includes one Base Station (BS) and multiple users is investigated. Both the uplink and downlink of the system apply the OFDM transmission. In the downlink, the BS transmits information and power over different subcarriers to the users simultaneously. In the uplink, the user transmits information to the BS by using the power harvested from the BS in the downlink. This paper aims to maximize the weighted sum of the downlink and uplink achievable rates by jointly optimizing subcarrier allocation and power allocation of the uplink and downlink. An optimal algorithm is proposed to solve this resulted optimization problem, which is based on the Lagrange duality method and the ellipsoid method. The performances of the proposed algorithm are verified by computer simulations.
2019, 41(6): 1365-1373.
doi: 10.11999/JEIT180157
Abstract:
To deal with the problem of joint estimation of spreading codes and information sequences for asynchronous long code DS-CDMA signals in multipath channels, an algorithm is introduced based on Sequential Monte Carlo(SMC) for blind estimation. The proposed algorithm emploies hybrid important sampling density to draw the samples from joint posterior distribution iteratively, and computes the importance weight to complete the estimation of the state variable. During the realization of the algorithm, in order to reduce the computational complexity, the modified algorithm estimates the spreading code of each user firstly, then processes the observation data, thereby modifies the original iteration step. Simulation results verify the adaptability of the proposed algorithms for multiple conditions. Moreover, it can obtain good estimation performance in time varying multipath channels.
To deal with the problem of joint estimation of spreading codes and information sequences for asynchronous long code DS-CDMA signals in multipath channels, an algorithm is introduced based on Sequential Monte Carlo(SMC) for blind estimation. The proposed algorithm emploies hybrid important sampling density to draw the samples from joint posterior distribution iteratively, and computes the importance weight to complete the estimation of the state variable. During the realization of the algorithm, in order to reduce the computational complexity, the modified algorithm estimates the spreading code of each user firstly, then processes the observation data, thereby modifies the original iteration step. Simulation results verify the adaptability of the proposed algorithms for multiple conditions. Moreover, it can obtain good estimation performance in time varying multipath channels.
2019, 41(6): 1374-1380.
doi: 10.11999/JEIT180261
Abstract:
In broadband Power-Line Communications (PLC), the background noise commonly assumed as Gaussian may not truly depict the effect of the human activities on noise characteristics. Symmetric Alpha-Stable (S\begin{document}$\alpha$\end{document} ![]()
![]()
\begin{document}$\alpha$\end{document} ![]()
![]()
\begin{document}$\alpha$\end{document} ![]()
![]()
\begin{document}$\alpha$\end{document} ![]()
![]()
In broadband Power-Line Communications (PLC), the background noise commonly assumed as Gaussian may not truly depict the effect of the human activities on noise characteristics. Symmetric Alpha-Stable (S
2019, 41(6): 1381-1388.
doi: 10.11999/JEIT180637
Abstract:
The traditional fingerprinting localization algorithm has high construct time overhead and low positioning accuracy. Because of this problem, an adaptive fading memory based bluetooth sequence matching localization algorithm is proposed. Firsly, Pedestrian Dead Reckoning(PDR) and Nearest Neighbor Algorithm(NNA) are applied to performing position calibration and Received Signal Strength(RSS) mapping of Motion Sequences. Secoudly, according to the relevance of neighboring locations, a sequence recursive search method is used to construct fingerprint sequence database. Finally, an adaptive fading memory algorithm and initial sequence matching degree are considered to realize the position estimation of target. The experimental results show that this algorithm is able to consume low construct time overhead and achieve high indoor localization precision.
The traditional fingerprinting localization algorithm has high construct time overhead and low positioning accuracy. Because of this problem, an adaptive fading memory based bluetooth sequence matching localization algorithm is proposed. Firsly, Pedestrian Dead Reckoning(PDR) and Nearest Neighbor Algorithm(NNA) are applied to performing position calibration and Received Signal Strength(RSS) mapping of Motion Sequences. Secoudly, according to the relevance of neighboring locations, a sequence recursive search method is used to construct fingerprint sequence database. Finally, an adaptive fading memory algorithm and initial sequence matching degree are considered to realize the position estimation of target. The experimental results show that this algorithm is able to consume low construct time overhead and achieve high indoor localization precision.
2019, 41(6): 1389-1396.
doi: 10.11999/JEIT180357
Abstract:
To maximize the long-term spectral efficiency and energy efficiency of a full duplex wireless access and backhaul integrated small base station scene, approximate dynamic programming based joint admission control and resource allocation optimization algorithm is proposed. The algorithm firstly considers the resource usage and power configuration of the current base station, the dynamic demand of user, the constraints of average delay as well as backhaul rate and transmission power. The corresponding multi-objective optimization model of maximum spectrum efficiency and minimizes power consumption is established by using the Constrained Markov Decision Process(CMDP). Then, the Chebyshev theory is used to transform the multi-objective into a single-objective optimization, and the Lagrange dual decomposition method is then used to convert the single-objective problem into unrestricted Markov decision process problem. Finally, To solve the " dimension disaster” explosion that generated when solving this unrestricted Markov Decision Process(MDP) problem, a dynamic resource allocation algorithms based on approximate dynamic programming is presented, and the access and resource allocation strategy is obtained during this process. The simulation results show that the algorithm can maximize the long-term average spectrum efficiency and energy efficiency, within the constraints of the average delay, backhaul rate and transmission power, under the scenario of integrated access and backhaul small base station.
To maximize the long-term spectral efficiency and energy efficiency of a full duplex wireless access and backhaul integrated small base station scene, approximate dynamic programming based joint admission control and resource allocation optimization algorithm is proposed. The algorithm firstly considers the resource usage and power configuration of the current base station, the dynamic demand of user, the constraints of average delay as well as backhaul rate and transmission power. The corresponding multi-objective optimization model of maximum spectrum efficiency and minimizes power consumption is established by using the Constrained Markov Decision Process(CMDP). Then, the Chebyshev theory is used to transform the multi-objective into a single-objective optimization, and the Lagrange dual decomposition method is then used to convert the single-objective problem into unrestricted Markov decision process problem. Finally, To solve the " dimension disaster” explosion that generated when solving this unrestricted Markov Decision Process(MDP) problem, a dynamic resource allocation algorithms based on approximate dynamic programming is presented, and the access and resource allocation strategy is obtained during this process. The simulation results show that the algorithm can maximize the long-term average spectrum efficiency and energy efficiency, within the constraints of the average delay, backhaul rate and transmission power, under the scenario of integrated access and backhaul small base station.
2019, 41(6): 1397-1404.
doi: 10.11999/JEIT180666
Abstract:
To solve the problem of real-time migration of Virtual Network Function (VNF) caused by lacking effective prediction in 5G network, a VNF migration algorithm based on deep belief network prediction of resource requirements is proposed. The algorithm builds firstly a system cost evaluation model integrating bandwidth cost and migration cost,and then designs a deep belief network prediction algorithm based on online learning which adopts adaptive learning rate and introduces multi-task learning mode to predict future resource requirements. Finally, based on the prediction result as well as the perception of network topology and resources, the VNFs are migrated to the physical nodes that meet the resource threshold constraints through greedy selection algorithm with the goal to optimize system cost,and then a migration mechanism based on tabu search is proposed to further optimize the migration strategy.The simulation results show that the prediction model can obtain good prediction results and adaptive learning rate accelerates the convergence speed of the training network.Moreover, the combination with the migration algorithm reduces effectively system cost and the number of Service Level Agreements (SLA) violations during the migration process, and improves the performance of network services.
To solve the problem of real-time migration of Virtual Network Function (VNF) caused by lacking effective prediction in 5G network, a VNF migration algorithm based on deep belief network prediction of resource requirements is proposed. The algorithm builds firstly a system cost evaluation model integrating bandwidth cost and migration cost,and then designs a deep belief network prediction algorithm based on online learning which adopts adaptive learning rate and introduces multi-task learning mode to predict future resource requirements. Finally, based on the prediction result as well as the perception of network topology and resources, the VNFs are migrated to the physical nodes that meet the resource threshold constraints through greedy selection algorithm with the goal to optimize system cost,and then a migration mechanism based on tabu search is proposed to further optimize the migration strategy.The simulation results show that the prediction model can obtain good prediction results and adaptive learning rate accelerates the convergence speed of the training network.Moreover, the combination with the migration algorithm reduces effectively system cost and the number of Service Level Agreements (SLA) violations during the migration process, and improves the performance of network services.
2019, 41(6): 1405-1412.
doi: 10.11999/JEIT181040
Abstract:
The secret key generation method based on random signal may leak part of the common randomness information and reduce the achievable secret key rate when legal transmitter transmits random signal. In response to this problem, the secret key generation method based on multi-stream random signal is proposed. Firstly, the transmitter uses the channel reciprocity and uplink pilot to estimate the downlink channel, then the transmitter transmits mutually independent signal on every antenna. The eavesdropper is difficult to estimate all the random signals. It is difficult to estimate all the random signals for the eavesdropper, so the overlapping signal received by every antenna is difficult to be obtained by the eavesdropper. However, the legal transmitter is able to calculate the signal received by legal receiver by using the downlink channel estimated and the signal transmitted. So, the overlapping signal on every legal antenna can be used to extract secret key as common randomness. Also, the achievable secret key rate expression and the mutual information expression of common randomness are derived, and the relationship between them and the secret key security is analyzed. At last, the effectiveness of this method is verified by the simulation. The simulation results show that this method can reduce the common randomness observed by the eavesdropper to raise the achievable secret key rate and secret key security.
The secret key generation method based on random signal may leak part of the common randomness information and reduce the achievable secret key rate when legal transmitter transmits random signal. In response to this problem, the secret key generation method based on multi-stream random signal is proposed. Firstly, the transmitter uses the channel reciprocity and uplink pilot to estimate the downlink channel, then the transmitter transmits mutually independent signal on every antenna. The eavesdropper is difficult to estimate all the random signals. It is difficult to estimate all the random signals for the eavesdropper, so the overlapping signal received by every antenna is difficult to be obtained by the eavesdropper. However, the legal transmitter is able to calculate the signal received by legal receiver by using the downlink channel estimated and the signal transmitted. So, the overlapping signal on every legal antenna can be used to extract secret key as common randomness. Also, the achievable secret key rate expression and the mutual information expression of common randomness are derived, and the relationship between them and the secret key security is analyzed. At last, the effectiveness of this method is verified by the simulation. The simulation results show that this method can reduce the common randomness observed by the eavesdropper to raise the achievable secret key rate and secret key security.
2019, 41(6): 1413-1419.
doi: 10.11999/JEIT180896
Abstract:
Focusing on the problem of information leakage in secret key agreement, combining information reconciliation and privacy amplification, a method based on Secure Polar Code (SPC) is proposed, which builds the bridge from the condition of Quantized Bit Error Rate (QBER) to the requirement of Secret Key Outage Probability (SKOP). Firstly, QBER is modeled as the Transmitted Bit Error Rate (TBER) of Additional White Gaussian Noise (AWGN) channel, so the advantage of QBER is converted to the advantage of AWGN channel; Then, the TBER of each polarized sub-channel is calculated by Gaussian approximation, and the upper and lower bounds of decoded bit error rate are also derived. Finally, the SPC is constructed based on generic algorithm and SKOP threshold. Simulation results show that the proposed method satisfies the requirement of SKOP and achieves higher secret key agreement efficiency, compared with Low Density Parity Check (LDPC)-based method.
Focusing on the problem of information leakage in secret key agreement, combining information reconciliation and privacy amplification, a method based on Secure Polar Code (SPC) is proposed, which builds the bridge from the condition of Quantized Bit Error Rate (QBER) to the requirement of Secret Key Outage Probability (SKOP). Firstly, QBER is modeled as the Transmitted Bit Error Rate (TBER) of Additional White Gaussian Noise (AWGN) channel, so the advantage of QBER is converted to the advantage of AWGN channel; Then, the TBER of each polarized sub-channel is calculated by Gaussian approximation, and the upper and lower bounds of decoded bit error rate are also derived. Finally, the SPC is constructed based on generic algorithm and SKOP threshold. Simulation results show that the proposed method satisfies the requirement of SKOP and achieves higher secret key agreement efficiency, compared with Low Density Parity Check (LDPC)-based method.
2019, 41(6): 1420-1426.
doi: 10.11999/JEIT180703
Abstract:
Two constructions of Gaussian integer Zero Correlation Zone (ZCZ) sequence set are researched. In Construction I, the method of zero padding is implemented on the ZCZ sequence set, and then the Gaussian integer ZCZ can be obtained by the filtering operation. Furthermore, the degree of the Gaussian integer ZCZ sequence set is calculated in this paper. In Construction II, two constructions of Gaussian integer orthogonal matrix are proposed. In addition, the optimal Gaussian integer ZCZ sequence sets are constructed based on the orthogonal matrix. The two classes of Gaussian integer ZCZ sequence sets presented in this paper can be applied to many communication systems such as Quasi-Synchronous Code Division Multiple Access (QS-CDMA), Orthogonal Frequency Division Multiplexing (OFDM) and Mutiple-Input Multiple-Output (MIMO) system to suppress the interference and improve the spectrum efficiency.
Two constructions of Gaussian integer Zero Correlation Zone (ZCZ) sequence set are researched. In Construction I, the method of zero padding is implemented on the ZCZ sequence set, and then the Gaussian integer ZCZ can be obtained by the filtering operation. Furthermore, the degree of the Gaussian integer ZCZ sequence set is calculated in this paper. In Construction II, two constructions of Gaussian integer orthogonal matrix are proposed. In addition, the optimal Gaussian integer ZCZ sequence sets are constructed based on the orthogonal matrix. The two classes of Gaussian integer ZCZ sequence sets presented in this paper can be applied to many communication systems such as Quasi-Synchronous Code Division Multiple Access (QS-CDMA), Orthogonal Frequency Division Multiplexing (OFDM) and Mutiple-Input Multiple-Output (MIMO) system to suppress the interference and improve the spectrum efficiency.
2019, 41(6): 1427-1433.
doi: 10.11999/JEIT180691
Abstract:
In order to improve the comprehensive performance of the wireless network intrusion detection model, Recurrent Neural Network (RNN) algorithm is used to build a wireless network intrusion detection classification model. For the over-fitting problem of the classification model caused by the imbalance of training data samples distribution in wireless network intrusion detection, based on the pre-treatment of raw data cleaning, transformation, feature selection, etc., an instance selection algorithm based on window is proposed to refine the train data-set. The network structure, activation function and re-usability of the attack classification model are optimized experimentally, so the optimization model is obtained finally. The classification accuracy of the optimization model is 98.6699%, and the running time after the model reuse optimization is 9.13 s. Compared to other machine learning algorithms, the proposed approach achieves good results in classification accuracy and execution efficiency. The comprehensive performances of the proposed model are better than those of traditional intrusion detection model.
In order to improve the comprehensive performance of the wireless network intrusion detection model, Recurrent Neural Network (RNN) algorithm is used to build a wireless network intrusion detection classification model. For the over-fitting problem of the classification model caused by the imbalance of training data samples distribution in wireless network intrusion detection, based on the pre-treatment of raw data cleaning, transformation, feature selection, etc., an instance selection algorithm based on window is proposed to refine the train data-set. The network structure, activation function and re-usability of the attack classification model are optimized experimentally, so the optimization model is obtained finally. The classification accuracy of the optimization model is 98.6699%, and the running time after the model reuse optimization is 9.13 s. Compared to other machine learning algorithms, the proposed approach achieves good results in classification accuracy and execution efficiency. The comprehensive performances of the proposed model are better than those of traditional intrusion detection model.
2019, 41(6): 1434-1441.
doi: 10.11999/JEIT180653
Abstract:
A Bayesian network structure learning algorithm based on improved whale optimization strategy is proposed to solve the problem that the current Bayesian network structure learning algorithm is easily trapped in local optimal and is of low optimization efficiency. The improved algorithm proposes first a new method to establish a better initial population, and then it uses the cross mutation operator that does not produce the illegal structure to construct an improved predation behavior suitable for Bayesian network structure learning. At the same time, it adopts the dynamic parameter tuning strategy to enhance the individual search ability. The population is updated followed by the fitness order so that the optimal Bayesian network structure is obtained. Simulation results demonstrate that the algorithm has global convergence, high efficiency and higher accuracy than other similar optimization algorithms.
A Bayesian network structure learning algorithm based on improved whale optimization strategy is proposed to solve the problem that the current Bayesian network structure learning algorithm is easily trapped in local optimal and is of low optimization efficiency. The improved algorithm proposes first a new method to establish a better initial population, and then it uses the cross mutation operator that does not produce the illegal structure to construct an improved predation behavior suitable for Bayesian network structure learning. At the same time, it adopts the dynamic parameter tuning strategy to enhance the individual search ability. The population is updated followed by the fitness order so that the optimal Bayesian network structure is obtained. Simulation results demonstrate that the algorithm has global convergence, high efficiency and higher accuracy than other similar optimization algorithms.
2019, 41(6): 1442-1449.
doi: 10.11999/JEIT180595
Abstract:
Semi-honest data collectors may cause privacy leaks during the collection and use of Sensitive Attribute (SA) data. In view of the problem, real-time data leaders are added in the traditional model and a privacy-protected data collection protocol based on the improved model is proposed. Without the assumption of trusted third party, the protocol ensures that data collectors maximization data utility can only be established on the basis of K-anonymized data. Data owners participates in the protocol flow in a distributed and collaborative manner to achieve the transmission of SA after the Quasi-Identifier (QI) is anonymized. This reduces the probability that the data collector uses the QI to associate SA values and weakens the risk of privacy leakage caused by internal identity disclosure. It divides the coded value of the SA into two shares of a random anchor point and a compensation distance through the tree coding structure and the members of the equivalent class formed by K-anonymity elect two data leaders to aggregate and forward the two shares respectively, which releases the association between unique network identification and SA values and prevents leakage of privacy caused by external identification effectively. Formal rules are established that meet the characteristics of the protocol and analyze the protocol to prove that the protocol meets privacy protection requirements.
Semi-honest data collectors may cause privacy leaks during the collection and use of Sensitive Attribute (SA) data. In view of the problem, real-time data leaders are added in the traditional model and a privacy-protected data collection protocol based on the improved model is proposed. Without the assumption of trusted third party, the protocol ensures that data collectors maximization data utility can only be established on the basis of K-anonymized data. Data owners participates in the protocol flow in a distributed and collaborative manner to achieve the transmission of SA after the Quasi-Identifier (QI) is anonymized. This reduces the probability that the data collector uses the QI to associate SA values and weakens the risk of privacy leakage caused by internal identity disclosure. It divides the coded value of the SA into two shares of a random anchor point and a compensation distance through the tree coding structure and the members of the equivalent class formed by K-anonymity elect two data leaders to aggregate and forward the two shares respectively, which releases the association between unique network identification and SA values and prevents leakage of privacy caused by external identification effectively. Formal rules are established that meet the characteristics of the protocol and analyze the protocol to prove that the protocol meets privacy protection requirements.
2019, 41(6): 1450-1457.
doi: 10.11999/JEIT180631
Abstract:
For problems of not meeting the demand of sampling both large flows and small flows at the same time, and not distinguishing flash crowd from traffic attacks in building network traffic anomaly detection datasets based on probabilistic sampling methods, a probabilistic flow sampling method for traffic anomaly detection is proposed. On the basis of the classification of network data flows according to their destination and source IP addresses, the sampling probability for each class of data flows is set as the maximum of its destination and source IP address’s sampling probability, and the number of sampled data flows is ceiled to ensure that each class of data flows is sampled at least once, so that the sampled dataset can reflect the distributions of large, small flows and source, destination IP addresses in original traffics. Then, the source IP address entropy is used to characterize the source IP dispersion of anomaly flows, and the attack flow sampling algorithm is designed based on the threshold of the source IP address entropy, which reduces the sampling probability of non-attack anomaly flows caused by flash crowd. The simulation results show that the proposed method can satisfy the sampling requirements of both large flows and small flows, it has a high anomaly flows sampling ability, can sample all the suspicious sources and destination IP addresses related to anomaly flows, and can effectively filter the non-attack anomaly flows.
For problems of not meeting the demand of sampling both large flows and small flows at the same time, and not distinguishing flash crowd from traffic attacks in building network traffic anomaly detection datasets based on probabilistic sampling methods, a probabilistic flow sampling method for traffic anomaly detection is proposed. On the basis of the classification of network data flows according to their destination and source IP addresses, the sampling probability for each class of data flows is set as the maximum of its destination and source IP address’s sampling probability, and the number of sampled data flows is ceiled to ensure that each class of data flows is sampled at least once, so that the sampled dataset can reflect the distributions of large, small flows and source, destination IP addresses in original traffics. Then, the source IP address entropy is used to characterize the source IP dispersion of anomaly flows, and the attack flow sampling algorithm is designed based on the threshold of the source IP address entropy, which reduces the sampling probability of non-attack anomaly flows caused by flash crowd. The simulation results show that the proposed method can satisfy the sampling requirements of both large flows and small flows, it has a high anomaly flows sampling ability, can sample all the suspicious sources and destination IP addresses related to anomaly flows, and can effectively filter the non-attack anomaly flows.
2019, 41(6): 1458-1465.
doi: 10.11999/JEIT180831
Abstract:
Long configuration time is a significant factor which restricts the performance improvement of the reconfigurable system, and a reasonable task scheduling technology can effectively reduce the system configuration time. A three-dimensional task scheduling model for Coarse-Grain Dynamic Reconfigurable System (CGDRS) and flow applications with data dependencies is proposed. Firstly, based on this model, a Configuration Prefetching Schedule Algorithm (CPSA) applying pre-configured strategy is designed. Then, the interval and continuous configuration reuse strategy are proposed according to the configuration reusability between tasks, and the CPSA algorithm is improved accordingly. The experimental results show this algorithm can avoid scheduling deadlock, reduce the execution time of flow applications and improve scheduling success rate. The optimization ratio of total execution time of flow applications achieves 6.13%～19.53% averagely compared with other scheduling algorithms.
Long configuration time is a significant factor which restricts the performance improvement of the reconfigurable system, and a reasonable task scheduling technology can effectively reduce the system configuration time. A three-dimensional task scheduling model for Coarse-Grain Dynamic Reconfigurable System (CGDRS) and flow applications with data dependencies is proposed. Firstly, based on this model, a Configuration Prefetching Schedule Algorithm (CPSA) applying pre-configured strategy is designed. Then, the interval and continuous configuration reuse strategy are proposed according to the configuration reusability between tasks, and the CPSA algorithm is improved accordingly. The experimental results show this algorithm can avoid scheduling deadlock, reduce the execution time of flow applications and improve scheduling success rate. The optimization ratio of total execution time of flow applications achieves 6.13%～19.53% averagely compared with other scheduling algorithms.
2019, 41(6): 1466-1471.
doi: 10.11999/JEIT180752
Abstract:
A 10 bit fully differential dual slope Analog-to-Digital Converter (ADC) for Time Delay Integration (TDI) CMOS image sensors is realized based on column-parallel single-slope ADC. Top plates of the two capacitors are used for sampling differential inputs, and the bottom plates are connected to ramp generator for conversion. Current steering is used to generate the rising and falling ramp with the same step voltage simultaneously. The proposed ADC is fabricated in SMIC 0.18 μm CMOS process. Simulated spurious free dynamic range and effective number of bits are 87.92 dB and 9.84 bit with the input frequency of 1.32 kHz at 19.49 kS/s sampling rate, respectively. Measured results show that the ADC has a differential nonlinearity of –0.7/+0.6 LSB and integral nonlinearity of –2.6/+2.1 LSB.
A 10 bit fully differential dual slope Analog-to-Digital Converter (ADC) for Time Delay Integration (TDI) CMOS image sensors is realized based on column-parallel single-slope ADC. Top plates of the two capacitors are used for sampling differential inputs, and the bottom plates are connected to ramp generator for conversion. Current steering is used to generate the rising and falling ramp with the same step voltage simultaneously. The proposed ADC is fabricated in SMIC 0.18 μm CMOS process. Simulated spurious free dynamic range and effective number of bits are 87.92 dB and 9.84 bit with the input frequency of 1.32 kHz at 19.49 kS/s sampling rate, respectively. Measured results show that the ADC has a differential nonlinearity of –0.7/+0.6 LSB and integral nonlinearity of –2.6/+2.1 LSB.
2019, 41(6): 1472-1479.
doi: 10.11999/JEIT180651
Abstract:
In order to deal with these issues of the traditional Fuzzy C-Means (FCM) algorithm, such as without consideration of the spatial neighborhood information of pixels, noise sensitivity and low convergence speed, a suppressed non-local spatial intuitionistic fuzzy c-means image segmentation algorithm is proposed. Firstly, in order to improve the accuracy of segmentation image, the non-local spatial information of pixel is used to improve anti-noise ability, and to overcome the shortcomings of the traditional FCM algorithm, which only considers the gray characteristic information of single pixel. Secondly, by using the ‘voting model’ based on the intuitionistic fuzzy set theory, the hesitation degrees are adaptively generated as inhibitory factors to modify the membership degrees, and then the operating efficiency is increased. Experimental results show that the new algorithm is robust to noise and has better segmentation performance.
In order to deal with these issues of the traditional Fuzzy C-Means (FCM) algorithm, such as without consideration of the spatial neighborhood information of pixels, noise sensitivity and low convergence speed, a suppressed non-local spatial intuitionistic fuzzy c-means image segmentation algorithm is proposed. Firstly, in order to improve the accuracy of segmentation image, the non-local spatial information of pixel is used to improve anti-noise ability, and to overcome the shortcomings of the traditional FCM algorithm, which only considers the gray characteristic information of single pixel. Secondly, by using the ‘voting model’ based on the intuitionistic fuzzy set theory, the hesitation degrees are adaptively generated as inhibitory factors to modify the membership degrees, and then the operating efficiency is increased. Experimental results show that the new algorithm is robust to noise and has better segmentation performance.
2019, 41(6): 1480-1487.
doi: 10.11999/JEIT180588
Abstract:
For the deficiency of traditional Continuously Adaptive Mean-shift (CAMshift) tracking algorithm can easily contain a large number of color information which belongs to the background in the process of establishing the target color model, an improved algorithm is proposed. The original image is divided into foreground and background based on the Gaussian Mixture Model(GMM). In the original image and the background image, the histogram of the hue component is established. Hue histograms of the background image are used to calculate the weight of the hue component in the original image. The hues belonging to the background are suppressed and the color differences between foreground and background are expanded. Experiment shows that by suppressing the hue components belonging to the background, the saliency of the target color model is expanded. The accuracy and stability of the target recognition are improved. The ratio of the max deviation to the target is less than 20%, which ensures the target not to be lost.
For the deficiency of traditional Continuously Adaptive Mean-shift (CAMshift) tracking algorithm can easily contain a large number of color information which belongs to the background in the process of establishing the target color model, an improved algorithm is proposed. The original image is divided into foreground and background based on the Gaussian Mixture Model(GMM). In the original image and the background image, the histogram of the hue component is established. Hue histograms of the background image are used to calculate the weight of the hue component in the original image. The hues belonging to the background are suppressed and the color differences between foreground and background are expanded. Experiment shows that by suppressing the hue components belonging to the background, the saliency of the target color model is expanded. The accuracy and stability of the target recognition are improved. The ratio of the max deviation to the target is less than 20%, which ensures the target not to be lost.
2019, 41(6): 1488-1495.
doi: 10.11999/JEIT180670
Abstract:
The diversity of the population and the crossover operator algorithm play an important role in solving global optimization problems in Differential Evolution (DE). The Multi-poplutions Covariance learning Differential Evolution (MCDE) algorithm is proposed. Firstly, the population structure is a multi-poplutions mechanism, and each subpopulation combines the corresponding mutation strategy to ensure the individual diversity in the evolutionary process. Then, the covariance learning establishes a proper rotation coordinate system for the crossover operation in the population. At the same time, the adaptive control parameters are used to balance the ability of population survey and convergence. Finally, the proposed algorithm is conducted on 25 benchmark functions including unimodal, multimodal, shifted and high-dimensional test functions and compared with the state-of-the-art evolutionary algorithms. The experimental results show that the proposed algorithm compared with other algorithms has the best effect on solving the global optimization problem.
The diversity of the population and the crossover operator algorithm play an important role in solving global optimization problems in Differential Evolution (DE). The Multi-poplutions Covariance learning Differential Evolution (MCDE) algorithm is proposed. Firstly, the population structure is a multi-poplutions mechanism, and each subpopulation combines the corresponding mutation strategy to ensure the individual diversity in the evolutionary process. Then, the covariance learning establishes a proper rotation coordinate system for the crossover operation in the population. At the same time, the adaptive control parameters are used to balance the ability of population survey and convergence. Finally, the proposed algorithm is conducted on 25 benchmark functions including unimodal, multimodal, shifted and high-dimensional test functions and compared with the state-of-the-art evolutionary algorithms. The experimental results show that the proposed algorithm compared with other algorithms has the best effect on solving the global optimization problem.
2019, 41(6): 1496-1502.
doi: 10.11999/JEIT180702
Abstract:
Because of the classic Faster RCNN training proccess with too many difficult training samples and low recall rate problem, a method which combines the techniques of Online Hard Example Mining (OHEM) and Hard Negative Example Mining (HNEM) is adopted, which carries out the error transfer for the difficult samples using its corresponding maximum loss value from real-time filtering. It solves the problem of low detection of hard example and improves the efficiency of the model training. To improve the recall rate and generalization of the model, an improved Non-Maximum Suppression (NMS) algorithm is proposed by setting confidence thresholds penalty function; In addition, multi-scale training and data augmentation are also introduced. Finally, the results before and after improvement are compared: Sensibility experiments show that the algorithm achieves good results in VOC2007 data set and VOC2012 data set, with the mean Average Percision (mAP) increasing from 69.9% to 74.40%, and 70.4% to 79.3% respectively, which demonstrates strongly the superiority of the algorithm.
Because of the classic Faster RCNN training proccess with too many difficult training samples and low recall rate problem, a method which combines the techniques of Online Hard Example Mining (OHEM) and Hard Negative Example Mining (HNEM) is adopted, which carries out the error transfer for the difficult samples using its corresponding maximum loss value from real-time filtering. It solves the problem of low detection of hard example and improves the efficiency of the model training. To improve the recall rate and generalization of the model, an improved Non-Maximum Suppression (NMS) algorithm is proposed by setting confidence thresholds penalty function; In addition, multi-scale training and data augmentation are also introduced. Finally, the results before and after improvement are compared: Sensibility experiments show that the algorithm achieves good results in VOC2007 data set and VOC2012 data set, with the mean Average Percision (mAP) increasing from 69.9% to 74.40%, and 70.4% to 79.3% respectively, which demonstrates strongly the superiority of the algorithm.
2019, 41(6): 1503-1509.
doi: 10.11999/JEIT180640
Abstract:
To solve the problem of insufficient number of participants and poor data quality in the sensing mission, a mobile crowd sensing incentive model for mission cost difference is proposed. First of all, the fuzzy reasoning method is used to analyze the impact of data quantity, environmental conditions and equipment consumption on mission cost, and the sensing mission is divided into different levels on the basis of cost difference. Meanwhile, the method is used to prepare a budget for the requester and give the participant an appropriate reward. Then, the sensing mission is assigned to more appropriate participants to complete the sensing mission and upload the sensing data through credibility assessment and participants’ preference. Finally, the sensing data uploaded by participants is evaluated, and the credibility of participants is updated. Besides, the participants are paid according to the cost level of perceived missions. The simulation experiments based on the real data set show that the model can recruit more users to participate in the sensing mission effectively and promote participants to upload high-quality sensing data by using the mutual influence between different modules.
To solve the problem of insufficient number of participants and poor data quality in the sensing mission, a mobile crowd sensing incentive model for mission cost difference is proposed. First of all, the fuzzy reasoning method is used to analyze the impact of data quantity, environmental conditions and equipment consumption on mission cost, and the sensing mission is divided into different levels on the basis of cost difference. Meanwhile, the method is used to prepare a budget for the requester and give the participant an appropriate reward. Then, the sensing mission is assigned to more appropriate participants to complete the sensing mission and upload the sensing data through credibility assessment and participants’ preference. Finally, the sensing data uploaded by participants is evaluated, and the credibility of participants is updated. Besides, the participants are paid according to the cost level of perceived missions. The simulation experiments based on the real data set show that the model can recruit more users to participate in the sensing mission effectively and promote participants to upload high-quality sensing data by using the mutual influence between different modules.
2019, 41(6): 1510-1517.
doi: 10.11999/JEIT180644
Abstract:
Nowadays, the civil aviation industry has a high-precision prediction demand of flight delays, thus a flight delay prediction model based on the deep SE-DenseNet is proposed. Firstly, flight data, associated airport delay information and meteorological data are fused in the model. Then, the improved SE-DenseNet algorithm is used to extract feature automatically based on the fused flight data set. Finally, the softmax classifier is used to predict the delay level of flight. The proposed SE-DenseNet, combing the advantages of DenseNet and SENet, can not only enhance the transmission of deep information, avoid the problem of vanishing gradients, but also achieve feature recalibration by the feature extraction process. The results indicate that after data fusion, the accuracy of the model is improved 1.8% than only considering the characteristics of the flight itself. The improved algorithm can effectively improve the network performance. The final accuracy of the model reaches 93.19%.
Nowadays, the civil aviation industry has a high-precision prediction demand of flight delays, thus a flight delay prediction model based on the deep SE-DenseNet is proposed. Firstly, flight data, associated airport delay information and meteorological data are fused in the model. Then, the improved SE-DenseNet algorithm is used to extract feature automatically based on the fused flight data set. Finally, the softmax classifier is used to predict the delay level of flight. The proposed SE-DenseNet, combing the advantages of DenseNet and SENet, can not only enhance the transmission of deep information, avoid the problem of vanishing gradients, but also achieve feature recalibration by the feature extraction process. The results indicate that after data fusion, the accuracy of the model is improved 1.8% than only considering the characteristics of the flight itself. The improved algorithm can effectively improve the network performance. The final accuracy of the model reaches 93.19%.
2019, 41(6): 1518-1524.
Abstract:
