Email alert
2022 Vol. 44, No. 10
Display Method:
2022, 44(10): 3343-3352.
doi: 10.11999/JEIT220380
Abstract:
Deep convolutional neural network-based image Super-Resolution (SR) methods assume generally that the degradations of Low-Resolution (LR) images are fixed and known (e.g., bicubic downsampling). Thus, they are almost unable to super-resolve images with unknown degradations (e.g., blur kernel and noise level). To address this problem, an iterative and alternative optimization-based blind image SR network is proposed, in which the blur kernel, noise level, and High-Resolution (HR) image are jointly estimated. Specifically, in the proposed method, the image reconstruction network reconstructs an HR image from the given LR image using the estimated blur kernel and noise level as prior knowledge. Correspondingly, the blur kernel and noise level estimators estimate the blur kernel and noise level respectively from the given LR image and the reconstructed HR image. To improve compatibility and promote each other mutually, the blur kernel estimator, noise level estimator, and image reconstruction network are iteratively and alternatively optimized in an end-to-end manner. The proposed network is compared with state-of-the-art methods (i.e., IKC, DASR, MANet, DAN) on commonly used benchmarks (i.e., Set5, Set14, B100, and Urban100) and real-world images. Results show that the proposed method achieves better performance on LR images with unknown degradations. Moreover, the proposed method has advantages in model size or processing efficiency.
Deep convolutional neural network-based image Super-Resolution (SR) methods assume generally that the degradations of Low-Resolution (LR) images are fixed and known (e.g., bicubic downsampling). Thus, they are almost unable to super-resolve images with unknown degradations (e.g., blur kernel and noise level). To address this problem, an iterative and alternative optimization-based blind image SR network is proposed, in which the blur kernel, noise level, and High-Resolution (HR) image are jointly estimated. Specifically, in the proposed method, the image reconstruction network reconstructs an HR image from the given LR image using the estimated blur kernel and noise level as prior knowledge. Correspondingly, the blur kernel and noise level estimators estimate the blur kernel and noise level respectively from the given LR image and the reconstructed HR image. To improve compatibility and promote each other mutually, the blur kernel estimator, noise level estimator, and image reconstruction network are iteratively and alternatively optimized in an end-to-end manner. The proposed network is compared with state-of-the-art methods (i.e., IKC, DASR, MANet, DAN) on commonly used benchmarks (i.e., Set5, Set14, B100, and Urban100) and real-world images. Results show that the proposed method achieves better performance on LR images with unknown degradations. Moreover, the proposed method has advantages in model size or processing efficiency.
2022, 44(10): 3353-3362.
doi: 10.11999/JEIT220375
Abstract:
Focusing on the serious color shift and loss of details caused by light absorption, backscattering and other factors in underwater images, an underwater image enhancement method based on multi-scale cascaded network is proposed in this paper. For the image gradient dissipation caused by incomplete utilization of features via single network, better details are preserved by cascading multi-scale original images and corresponding feature images, and rapid prediction of residuals from shallower layers to deeper layers can be realized at the same time. In addition, joint dense network block and recursive block are introduced to avoid effectively the problem of excessive parameters introduced by conventional multi-scale network through feature reuse. A joint loss function of Charbonnier and the Structural SIMilarity (SSIM) is proposed to solve effectively the problem of uneven restoration of image details caused by a single loss. The simulation experiments show that the proposed network has achieved excellent results in dealing with severe color shift and loss of details.
Focusing on the serious color shift and loss of details caused by light absorption, backscattering and other factors in underwater images, an underwater image enhancement method based on multi-scale cascaded network is proposed in this paper. For the image gradient dissipation caused by incomplete utilization of features via single network, better details are preserved by cascading multi-scale original images and corresponding feature images, and rapid prediction of residuals from shallower layers to deeper layers can be realized at the same time. In addition, joint dense network block and recursive block are introduced to avoid effectively the problem of excessive parameters introduced by conventional multi-scale network through feature reuse. A joint loss function of Charbonnier and the Structural SIMilarity (SSIM) is proposed to solve effectively the problem of uneven restoration of image details caused by a single loss. The simulation experiments show that the proposed network has achieved excellent results in dealing with severe color shift and loss of details.
2022, 44(10): 3363-3371.
doi: 10.11999/JEIT211012
Abstract:
The underwater image quality is seriously degraded due to the effects of absorption and scattering when light propagates underwater. In order to remove color distortion and blur, and improve the quality of underwater image effectively, an underwater image restoration method based on background light corrected image formation model is proposed in this paper. Based on the observation of ground hazy images, the assumption of background light offset for underwater images is put forward, which is the cornerstone of the background light corrected image formation model. Then, a monocular depth estimation network is used to obtain the estimate of the scene depth. Combined with the background light corrected image formation model, the underwater offset component is obtained by non-linear least square fitting, so as to remove water from underwater images. Finally, the transmittance of hazy image after water removed is optimized and combined with the corrected background light to achieve image recovery. Experimental results show that the method works well in restoring the original color of underwater scenes and removing scattered light.
The underwater image quality is seriously degraded due to the effects of absorption and scattering when light propagates underwater. In order to remove color distortion and blur, and improve the quality of underwater image effectively, an underwater image restoration method based on background light corrected image formation model is proposed in this paper. Based on the observation of ground hazy images, the assumption of background light offset for underwater images is put forward, which is the cornerstone of the background light corrected image formation model. Then, a monocular depth estimation network is used to obtain the estimate of the scene depth. Combined with the background light corrected image formation model, the underwater offset component is obtained by non-linear least square fitting, so as to remove water from underwater images. Finally, the transmittance of hazy image after water removed is optimized and combined with the corrected background light to achieve image recovery. Experimental results show that the method works well in restoring the original color of underwater scenes and removing scattered light.
2022, 44(10): 3372-3378.
doi: 10.11999/JEIT210761
Abstract:
In order to solve the problem of low detection accuracy of SSD-MV2, a Selective and Efficient Block (SEB) and a Selective and Deformable Block (SDB) are proposed. At the same time, the basic network and additional feature extraction network of SSD-MV2 are redesigned by using SEB and SDB, which is named SSD-MV2SDB, and a set of reasonable expansion coefficient of basic network and number of SDB in additional feature extraction network are selected for SSD-MV2SDB. On UOI-DET, mAP of SSD-MV2SDB is 3.04% higher than that of SSD-MV2. The experimental results show that SSD-MV2SDB is suitable for underwater optical image interested object detection task.
In order to solve the problem of low detection accuracy of SSD-MV2, a Selective and Efficient Block (SEB) and a Selective and Deformable Block (SDB) are proposed. At the same time, the basic network and additional feature extraction network of SSD-MV2 are redesigned by using SEB and SDB, which is named SSD-MV2SDB, and a set of reasonable expansion coefficient of basic network and number of SDB in additional feature extraction network are selected for SSD-MV2SDB. On UOI-DET, mAP of SSD-MV2SDB is 3.04% higher than that of SSD-MV2. The experimental results show that SSD-MV2SDB is suitable for underwater optical image interested object detection task.
2022, 44(10): 3379-3388.
doi: 10.11999/JEIT211593
Abstract:
In this paper, a novel zero-shot low-light image enhancement framework is proposed based on dual iterations. The outer iteration uses a network to estimate the enhancement parameters, with which the inner iteration improves actually the image, and the results are applied to calculating the loss functions and updating the outer network. After multiple rounds of iterations, high-quality images can be obtained. Within this framework, an adaptive parameter estimation module and an attention-based pixel-wise atmosphere estimation module are designed. In addition, unsupervised loss functions based on light, contrast, color balance and image smoothness priors are proposed. Experiments demonstrate that the proposed method obtains high quality clear images from low-light ones, and outperforms state-of-the-art methods. Furthermore, the proposed method belongs to zero-shot learning, which does not need training dataset and thus can be widely applied.
In this paper, a novel zero-shot low-light image enhancement framework is proposed based on dual iterations. The outer iteration uses a network to estimate the enhancement parameters, with which the inner iteration improves actually the image, and the results are applied to calculating the loss functions and updating the outer network. After multiple rounds of iterations, high-quality images can be obtained. Within this framework, an adaptive parameter estimation module and an attention-based pixel-wise atmosphere estimation module are designed. In addition, unsupervised loss functions based on light, contrast, color balance and image smoothness priors are proposed. Experiments demonstrate that the proposed method obtains high quality clear images from low-light ones, and outperforms state-of-the-art methods. Furthermore, the proposed method belongs to zero-shot learning, which does not need training dataset and thus can be widely applied.
2022, 44(10): 3389-3398.
doi: 10.11999/JEIT211517
Abstract:
The absorption and scattering properties of the water medium cause different types of distortion in underwater images, which affects seriously the accuracy and effectiveness of subsequent processing. At present, underwater image enhancement methods with supervised learning rely on synthetic underwater paired image sets for training. However, the supervised learning methods are challenging to apply to practical application scenarios because the synthetic data may not accurately model the underlying physical mechanisms of underwater imaging. An unsupervised underwater image enhancement based on feature disentanglement is proposed. On the one hand, considering the difficulty and high cost of acquiring clear-unclear paired datasets in the same scene, a cycle generative adversarial network is employed to convert the underwater image enhancement problem into a style transfer problem to achieve unsupervised learning. On the other hand, the feature disentanglement method is combined to extract the style features and structure features separately to ensure the structural consistency of the images before and after enhancement. The experimental results show that the method can effectively recover the color and texture details of underwater images in the case of unpaired data training.
The absorption and scattering properties of the water medium cause different types of distortion in underwater images, which affects seriously the accuracy and effectiveness of subsequent processing. At present, underwater image enhancement methods with supervised learning rely on synthetic underwater paired image sets for training. However, the supervised learning methods are challenging to apply to practical application scenarios because the synthetic data may not accurately model the underlying physical mechanisms of underwater imaging. An unsupervised underwater image enhancement based on feature disentanglement is proposed. On the one hand, considering the difficulty and high cost of acquiring clear-unclear paired datasets in the same scene, a cycle generative adversarial network is employed to convert the underwater image enhancement problem into a style transfer problem to achieve unsupervised learning. On the other hand, the feature disentanglement method is combined to extract the style features and structure features separately to ensure the structural consistency of the images before and after enhancement. The experimental results show that the method can effectively recover the color and texture details of underwater images in the case of unpaired data training.
2022, 44(10): 3399-3408.
doi: 10.11999/JEIT211131
Abstract:
Considering the difficult problems of brightness enhancement, noise suppression and maintaining texture color consistency in the low-light image enhancement model, a low-light image enhancement method based on the shifted window self-attention mechanism is proposed. Based on the U-shaped structure and the multi-head self-attention model of shifted windows, an image enhancement network composed of encoders, decoders and jump connections is constructed. The feature extraction advantages of the self-attention mechanism are applied to the field of low-light image enhancement and long-term dependence between image feature information is established, which can obtain global features effectively. The proposed method is compared width current popular algorithms in quantitative and qualitative comparison experiments, subjectively, the brightness of the image and noise suppression are significantly improved, and simultaneously better maintains the color information that highlights the texture details by the proposed method. In terms of objective indicators such as Peak Signal-to-Noise Ratio(PSNR), Structural SIMilarity index(SSIM), and Learned Perceptual Image Patch Similarity (LPIPS), which are improved 0.35 dB, 0.041 and 0.031 respectively compared with the optimal values of other methods. The experimental results show that the subjective perception quality and objective evaluation indicators of low-light images can be effectively improved by the proposed method, indicating a certain application value
Considering the difficult problems of brightness enhancement, noise suppression and maintaining texture color consistency in the low-light image enhancement model, a low-light image enhancement method based on the shifted window self-attention mechanism is proposed. Based on the U-shaped structure and the multi-head self-attention model of shifted windows, an image enhancement network composed of encoders, decoders and jump connections is constructed. The feature extraction advantages of the self-attention mechanism are applied to the field of low-light image enhancement and long-term dependence between image feature information is established, which can obtain global features effectively. The proposed method is compared width current popular algorithms in quantitative and qualitative comparison experiments, subjectively, the brightness of the image and noise suppression are significantly improved, and simultaneously better maintains the color information that highlights the texture details by the proposed method. In terms of objective indicators such as Peak Signal-to-Noise Ratio(PSNR), Structural SIMilarity index(SSIM), and Learned Perceptual Image Patch Similarity (LPIPS), which are improved 0.35 dB, 0.041 and 0.031 respectively compared with the optimal values of other methods. The experimental results show that the subjective perception quality and objective evaluation indicators of low-light images can be effectively improved by the proposed method, indicating a certain application value
2022, 44(10): 3409-3418.
doi: 10.11999/JEIT220381
Abstract:
Previous dehazing models trained on synthetic hazy images can not generalize well on real hazy scenes and improve the performance of high-level vision tasks significantly. To resolve this issue, a semi-supervised image dehazing based on multi-priors constrain and output consistency regularization is proposed. The algorithm adopts the encoder and decoder network to train on the synthetic and real hazy images by sharing the parameters. Multi prior-based dehazed images are adopted as pseudo labels to constrain the real scene hazy images. Furthermore, to reduce the divergence of different prior-based methods, the dehazing results of the random mix-up real hazy images are regularized to be consistent with the corresponding mix-up of the prior-based dehazed images. Finally, the experiment results demonstrate the performance of the proposed algorithm compared with the state-of-the-art methods.
Previous dehazing models trained on synthetic hazy images can not generalize well on real hazy scenes and improve the performance of high-level vision tasks significantly. To resolve this issue, a semi-supervised image dehazing based on multi-priors constrain and output consistency regularization is proposed. The algorithm adopts the encoder and decoder network to train on the synthetic and real hazy images by sharing the parameters. Multi prior-based dehazed images are adopted as pseudo labels to constrain the real scene hazy images. Furthermore, to reduce the divergence of different prior-based methods, the dehazing results of the random mix-up real hazy images are regularized to be consistent with the corresponding mix-up of the prior-based dehazed images. Finally, the experiment results demonstrate the performance of the proposed algorithm compared with the state-of-the-art methods.
2022, 44(10): 3419-3426.
doi: 10.11999/JEIT220260
Abstract:
Applying the object detection framework to the processing of underwater sonar images is a recent high-profile topic. Existing detection methods for sonar data are mainly based on the texture of sonar image. These methods are not able to handle the unstable geometric shape of objects in sonar image. To this end, a YOLOv3-based underwater object detection model YOLOv3F is proposed, which fuses texture features extracted from sonar images with spatial geometric features extracted from point clouds, and then the fused features are used for target detection. The experimental results show that the performance of the proposed improved model is significantly improved compared with the three control models in the experimental setup when different intersection of detection frames are set and compared to the detection of targets of different scales; The improved model also shows better detection results compared with YOLOv3 in the case of detection of a single class of objects.
Applying the object detection framework to the processing of underwater sonar images is a recent high-profile topic. Existing detection methods for sonar data are mainly based on the texture of sonar image. These methods are not able to handle the unstable geometric shape of objects in sonar image. To this end, a YOLOv3-based underwater object detection model YOLOv3F is proposed, which fuses texture features extracted from sonar images with spatial geometric features extracted from point clouds, and then the fused features are used for target detection. The experimental results show that the performance of the proposed improved model is significantly improved compared with the three control models in the experimental setup when different intersection of detection frames are set and compared to the detection of targets of different scales; The improved model also shows better detection results compared with YOLOv3 in the case of detection of a single class of objects.
2022, 44(10): 3427-3434.
doi: 10.11999/JEIT211324
Abstract:
Most image dehazing algorithms perform well in one or several homogeneous hazy map datasets, but process poor performance in datasets with different styles or nonhomogeneous hazy map datasets. Meanwhile, in practical application, the algorithm will be limited in model scenes due to poor model generalization ability. In view of the above situation, a Convolutional Neural Network (CNN) based on transfer learning is proposed to alleviate problems such as nonhomogeneous hazy map dehazing and defective generalization ability. ImageNet pre-trained model parameters are utilized as the initial parameters of the transfer learning model. In order to accelerate the convergence rate of model training, the algorithm is able to adapt quickly to different datasets. The model is composed of three subnets: residual feature sub network, local network and the overall feature extraction of feature extraction sub network. The model is ensured by the combination of three subnets to extract features from both the whole image feature and the local image feature, and achieves excellent dehazing effect in real hazy scenes (homogeneous and nonhomogeneous haze). In summary, the proposed method improves the model efficiency, haze removal quality and convenience of hazy map scene selection. To quantitatively and qualitatively measure the performance of the model, experiments are performed on NTIRE2020 and NTIRE2021, which are nonhomogeneous hazy map datasets with high haze removal difficulty. Experimental results show that the three-subnets model achieves outstanding results in both subjective and objective evaluation metrics. Unsatisfactory generalization performance of the algorithm and training difficulty are improved in small datasets. The architecture of three subnets can be widely utilized in small-scale datasets and changeable scene image dehazing projects.
Most image dehazing algorithms perform well in one or several homogeneous hazy map datasets, but process poor performance in datasets with different styles or nonhomogeneous hazy map datasets. Meanwhile, in practical application, the algorithm will be limited in model scenes due to poor model generalization ability. In view of the above situation, a Convolutional Neural Network (CNN) based on transfer learning is proposed to alleviate problems such as nonhomogeneous hazy map dehazing and defective generalization ability. ImageNet pre-trained model parameters are utilized as the initial parameters of the transfer learning model. In order to accelerate the convergence rate of model training, the algorithm is able to adapt quickly to different datasets. The model is composed of three subnets: residual feature sub network, local network and the overall feature extraction of feature extraction sub network. The model is ensured by the combination of three subnets to extract features from both the whole image feature and the local image feature, and achieves excellent dehazing effect in real hazy scenes (homogeneous and nonhomogeneous haze). In summary, the proposed method improves the model efficiency, haze removal quality and convenience of hazy map scene selection. To quantitatively and qualitatively measure the performance of the model, experiments are performed on NTIRE2020 and NTIRE2021, which are nonhomogeneous hazy map datasets with high haze removal difficulty. Experimental results show that the three-subnets model achieves outstanding results in both subjective and objective evaluation metrics. Unsatisfactory generalization performance of the algorithm and training difficulty are improved in small datasets. The architecture of three subnets can be widely utilized in small-scale datasets and changeable scene image dehazing projects.
2022, 44(10): 3435-3446.
doi: 10.11999/JEIT210787
Abstract:
In the research and development of high-end intelligent security check system, it is a challenging key technology how to make the detection of whether the human body is carrying hiding contraband quickly and efficiently in the normal process of non-contact travel. Passive millimeter wave imaging has become a popular option for security imaging due to its outstanding advantages such as safety, harmlessness and strong penetration. In this paper, the complementary advantages of passive millimeter wave imaging and visible imaging are employed, and a high-performance detection algorithm for hiding contraband in human body based on the lightweight U-Net is proposed. A lightweight U-Net is first constructed and trained to realize the rapid segmentation of the human contour in Passive MilliMeter Wave Image (PMMWI) and Visible Image (VI). In this way, the information of human contour and hiding contraband can be extracted. Then, human contour registration on PMMWI/VI is realized by the unsupervised learning method based on the similarity measure with the lightweight U-Net. After filtering the false alarm target, the position of the hiding contraband is marked in VI and the detection result on single frame image can be obtained. In the end, the final detection result is given through the comprehensive integration and inference of the detection results of sequence multi-frame images. Experimental results on a specially constructed dataset show that the proposed method reaches 92.3% of F1 evaluation index, thus demonstrates its performance advantages.
In the research and development of high-end intelligent security check system, it is a challenging key technology how to make the detection of whether the human body is carrying hiding contraband quickly and efficiently in the normal process of non-contact travel. Passive millimeter wave imaging has become a popular option for security imaging due to its outstanding advantages such as safety, harmlessness and strong penetration. In this paper, the complementary advantages of passive millimeter wave imaging and visible imaging are employed, and a high-performance detection algorithm for hiding contraband in human body based on the lightweight U-Net is proposed. A lightweight U-Net is first constructed and trained to realize the rapid segmentation of the human contour in Passive MilliMeter Wave Image (PMMWI) and Visible Image (VI). In this way, the information of human contour and hiding contraband can be extracted. Then, human contour registration on PMMWI/VI is realized by the unsupervised learning method based on the similarity measure with the lightweight U-Net. After filtering the false alarm target, the position of the hiding contraband is marked in VI and the detection result on single frame image can be obtained. In the end, the final detection result is given through the comprehensive integration and inference of the detection results of sequence multi-frame images. Experimental results on a specially constructed dataset show that the proposed method reaches 92.3% of F1 evaluation index, thus demonstrates its performance advantages.
2022, 44(10): 3447-3457.
doi: 10.11999/JEIT210764
Abstract:
ElectroEncephaloGraphy (EEG) is an important brain functional imaging technology. The task to reconstruct cortical activities based on the scalp EEG is called EEG source imaging. However, the accurate reconstruction of the locations and sizes of brain source activity remains a challenge. To employ fully the spatiotemporal complementary information of EEG and functional Magnetic Resonance Imaging (fMRI), a new EEG source imaging algorithm, i.e., FN-STCSI (Functional Network based Spatio-Temporal Constrains Source Imaging) is proposed. Specifically, to make full use of the temporal information of EEG signals, the source signal matrix is decomposed into a linear combination of several time basis functions based on the idea of matrix decomposition. Additionally, to fuse the high spatial resolution information of fMRI, FN-STCSI employes independent component analysis to extract the fMRI functional networks. Then these fMRI networks are used to construct the spatial covariance basis for EEG source imaging. Variational Bayesian inference techniques are used to determine the relative contribution of each spatial covariance basis to realize EEG-fMRI fusion. Through Monte Carlo numerical simulation and experimental data analysis, FN-STCSI is compared with existing algorithms under different signal-to-noise ratios and different prior conditions. The results show that FN-STCSI can effectively fuse the complementary spatiotemporal information of EEG-fMRI and improve the performance of EEG extended source imaging.
ElectroEncephaloGraphy (EEG) is an important brain functional imaging technology. The task to reconstruct cortical activities based on the scalp EEG is called EEG source imaging. However, the accurate reconstruction of the locations and sizes of brain source activity remains a challenge. To employ fully the spatiotemporal complementary information of EEG and functional Magnetic Resonance Imaging (fMRI), a new EEG source imaging algorithm, i.e., FN-STCSI (Functional Network based Spatio-Temporal Constrains Source Imaging) is proposed. Specifically, to make full use of the temporal information of EEG signals, the source signal matrix is decomposed into a linear combination of several time basis functions based on the idea of matrix decomposition. Additionally, to fuse the high spatial resolution information of fMRI, FN-STCSI employes independent component analysis to extract the fMRI functional networks. Then these fMRI networks are used to construct the spatial covariance basis for EEG source imaging. Variational Bayesian inference techniques are used to determine the relative contribution of each spatial covariance basis to realize EEG-fMRI fusion. Through Monte Carlo numerical simulation and experimental data analysis, FN-STCSI is compared with existing algorithms under different signal-to-noise ratios and different prior conditions. The results show that FN-STCSI can effectively fuse the complementary spatiotemporal information of EEG-fMRI and improve the performance of EEG extended source imaging.
2022, 44(10): 3458-3464.
doi: 10.11999/JEIT210323
Abstract:
As a classic feature fusion method, Canonical Correlation Analysis (CCA) is widely used in the field of pattern recognition. Its goal is to learn the relevant projection direction to maximize the correlation between the two sets of variables, but the class label information of the sample and the information between samples redundancy are not considered, which affects the supervisory sensitivity and discriminative power of the fused features. To this end, a label-sensitive Multi-set Discriminant Orthogonal Canonical Correlation Analysis (MDOCCA) feature fusion method is proposed. This method is based on canonical correlation analysis theory. The class label information is embedded into the feature fusion framework, and the orthogonal constraint is added to ensure the maximum fusion of features. Irrelevant, the redundancy of feature information is reduced and the discrimination is improved. Some experiments on multiple image data sets show that this method is an effective feature fusion method.
As a classic feature fusion method, Canonical Correlation Analysis (CCA) is widely used in the field of pattern recognition. Its goal is to learn the relevant projection direction to maximize the correlation between the two sets of variables, but the class label information of the sample and the information between samples redundancy are not considered, which affects the supervisory sensitivity and discriminative power of the fused features. To this end, a label-sensitive Multi-set Discriminant Orthogonal Canonical Correlation Analysis (MDOCCA) feature fusion method is proposed. This method is based on canonical correlation analysis theory. The class label information is embedded into the feature fusion framework, and the orthogonal constraint is added to ensure the maximum fusion of features. Irrelevant, the redundancy of feature information is reduced and the discrimination is improved. Some experiments on multiple image data sets show that this method is an effective feature fusion method.
2022, 44(10): 3465-3475.
doi: 10.11999/JEIT210716
Abstract:
Trajectory prediction is one of the core tasks in automatic driving system. At present, trajectory prediction algorithms based on deep learning involve information representation, perception and motion reasoning of targets. Considering the problem that the existing trajectory prediction models does not take into account the social motivation of pedestrians and can not effectively predict the local destination of pedestrians in different social conditions in the scene, a Conditional Endpoint local destination Pooling NETwork (CEPNET) is proposed. The network uses conditional variational autoencoder to map out the potential distribution in space, which can study the observation of the history track probability distribution in the specific scene. And then a local feature inference algorithm is built to code the similarity features of conditional endpoint as local destination features. Finally, the interference signals in the scene are filtered out by social pooling network. At the same time, self-attention social mask is used to enhance pedestrian’s self-attention. In order to verify the reliability of each module of the algorithm, the public datasets of Walking pedestrians In busy scenarios from a BIrd eye view(BIWI) and University of CYprus multi-person trajectory (UCY) are used to conduct ablation experiments, and compared with advanced trajectory prediction algorithms such as Vanilla, Socially acceptable trajectories with Generative Adversarial Networks (SGAN) and multimodal Trajectory forecasting using Bicycle-GAN and Graph Attention networks(S-BiGAT). The experimental results on the Trajnet++ benchmark show that compared with the benchmark Vanilla algorithm, the Average Displacement Error (ADE) is reduced by 22.52%, the Final Displacement Error (FDE) is reduced by 20%, the predicted collision rate Col-I is reduced by 9.75%, and the true collision rate Col-II is reduced by 9.15%.
Trajectory prediction is one of the core tasks in automatic driving system. At present, trajectory prediction algorithms based on deep learning involve information representation, perception and motion reasoning of targets. Considering the problem that the existing trajectory prediction models does not take into account the social motivation of pedestrians and can not effectively predict the local destination of pedestrians in different social conditions in the scene, a Conditional Endpoint local destination Pooling NETwork (CEPNET) is proposed. The network uses conditional variational autoencoder to map out the potential distribution in space, which can study the observation of the history track probability distribution in the specific scene. And then a local feature inference algorithm is built to code the similarity features of conditional endpoint as local destination features. Finally, the interference signals in the scene are filtered out by social pooling network. At the same time, self-attention social mask is used to enhance pedestrian’s self-attention. In order to verify the reliability of each module of the algorithm, the public datasets of Walking pedestrians In busy scenarios from a BIrd eye view(BIWI) and University of CYprus multi-person trajectory (UCY) are used to conduct ablation experiments, and compared with advanced trajectory prediction algorithms such as Vanilla, Socially acceptable trajectories with Generative Adversarial Networks (SGAN) and multimodal Trajectory forecasting using Bicycle-GAN and Graph Attention networks(S-BiGAT). The experimental results on the Trajnet++ benchmark show that compared with the benchmark Vanilla algorithm, the Average Displacement Error (ADE) is reduced by 22.52%, the Final Displacement Error (FDE) is reduced by 20%, the predicted collision rate Col-I is reduced by 9.75%, and the true collision rate Col-II is reduced by 9.15%.
2022, 44(10): 3476-3484.
doi: 10.11999/JEIT210812
Abstract:
An accurate relative localization is critical for multiple robots to realize collaboration and formation control. Visual or Light Detection And Ranging (LiDAR)-based approaches use feature matching to determine the relative pose between robots in indoor environments with Global Positioning System (GPS)-denied, but which is challenging in non-line-of-sight environment. To solve this problem, a relative positioning approach of mobile robots based on multiple Ultra WideBand (UWB) nodes is proposed. First, multiple UWB nodes carried by mobile robot are used to form UWB array, and the relative pose estimation between robots is realized through nonlinear optimization algorithm. To improve further the localization accuracy, the results of non-linear optimization are constrained through odometry measurements. In addition, in order to meet the real-time requirement, the relative pose and the odometry in the sliding window are optimized through the graph optimization algorithm. However, the uncertainty of the relative pose from the non-linear optimization is not known, thus it will affect the optimization accuracy. Therefore, this paper uses particle filtering to integrate the odometry and relative pose from sliding window to improve further the accuracy. The experimental results show that the proposed approach provides an average positioning error of 0.312 m and orientation error of 4.903° in an indoor environment with a size of 12×6 m, and has good real-time performance.
An accurate relative localization is critical for multiple robots to realize collaboration and formation control. Visual or Light Detection And Ranging (LiDAR)-based approaches use feature matching to determine the relative pose between robots in indoor environments with Global Positioning System (GPS)-denied, but which is challenging in non-line-of-sight environment. To solve this problem, a relative positioning approach of mobile robots based on multiple Ultra WideBand (UWB) nodes is proposed. First, multiple UWB nodes carried by mobile robot are used to form UWB array, and the relative pose estimation between robots is realized through nonlinear optimization algorithm. To improve further the localization accuracy, the results of non-linear optimization are constrained through odometry measurements. In addition, in order to meet the real-time requirement, the relative pose and the odometry in the sliding window are optimized through the graph optimization algorithm. However, the uncertainty of the relative pose from the non-linear optimization is not known, thus it will affect the optimization accuracy. Therefore, this paper uses particle filtering to integrate the odometry and relative pose from sliding window to improve further the accuracy. The experimental results show that the proposed approach provides an average positioning error of 0.312 m and orientation error of 4.903° in an indoor environment with a size of 12×6 m, and has good real-time performance.
2022, 44(10): 3485-3491.
doi: 10.11999/JEIT210729
Abstract:
This paper focused on the problem of the multi-access transmission in multiuser Multiple-Input Multiple-Output (MIMO) relay systems. In order to improve the performance in terms of system capacity and bit error rate, a joint precoding approach for the base station and the relay station is proposed by exploiting nested lattice coding which is capable of achieving modulo-channel capacity with additive Gaussian noise. At base station, the nested lattice coding is equipped up with Vector Perturbation (VP) precoding, leading to a reduction in transmit power. Since the modulo operation with respect to VP precoding is identical to that in nested coding, the relay can not only depress the noise power but also remove the redundant information with one-step modulo operation. An integer-forcing precoding method is designed for the relay to reduce the equivalent noise. Moreover, the acquisition of both the integral coefficient matrix and the vector perturbation matrix is provided. The transmit power allocation between the base station and the relay is optimized to maximize the system sum rate. The outstanding performance of the proposed scheme is validated by the simulation results.
This paper focused on the problem of the multi-access transmission in multiuser Multiple-Input Multiple-Output (MIMO) relay systems. In order to improve the performance in terms of system capacity and bit error rate, a joint precoding approach for the base station and the relay station is proposed by exploiting nested lattice coding which is capable of achieving modulo-channel capacity with additive Gaussian noise. At base station, the nested lattice coding is equipped up with Vector Perturbation (VP) precoding, leading to a reduction in transmit power. Since the modulo operation with respect to VP precoding is identical to that in nested coding, the relay can not only depress the noise power but also remove the redundant information with one-step modulo operation. An integer-forcing precoding method is designed for the relay to reduce the equivalent noise. Moreover, the acquisition of both the integral coefficient matrix and the vector perturbation matrix is provided. The transmit power allocation between the base station and the relay is optimized to maximize the system sum rate. The outstanding performance of the proposed scheme is validated by the simulation results.
2022, 44(10): 3492-3498.
doi: 10.11999/JEIT210772
Abstract:
To improve the operation cycle and energy utilization of Internet of Things (IoT) nodes, an energy-efficient maximization resource allocation algorithm is proposed for a multi-tag wireless-powered backscatter communication network. Specifically, a resource allocation model is developed to maximize the system energy efficiency under the transmission rate constraints, energy harvesting constraints, and transmit power constraints. The original fractional non-convex problem is transformed into a solvable convex optimization problem by using Dinkelbach's theory, a quadratic transformation method, and the variable substitution method. The globally optimal solutions of the considered problem are obtained by using Lagrange dual theory. Simulation results show that the proposed algorithm has better convergence and energy efficiency.
To improve the operation cycle and energy utilization of Internet of Things (IoT) nodes, an energy-efficient maximization resource allocation algorithm is proposed for a multi-tag wireless-powered backscatter communication network. Specifically, a resource allocation model is developed to maximize the system energy efficiency under the transmission rate constraints, energy harvesting constraints, and transmit power constraints. The original fractional non-convex problem is transformed into a solvable convex optimization problem by using Dinkelbach's theory, a quadratic transformation method, and the variable substitution method. The globally optimal solutions of the considered problem are obtained by using Lagrange dual theory. Simulation results show that the proposed algorithm has better convergence and energy efficiency.
2022, 44(10): 3499-3506.
doi: 10.11999/JEIT210768
Abstract:
In order to tackle the problem of single-mixed signal modulation type recognition with low efficiency and poor accuracy in satellite communication, based on clustering characteristics of constellation and high order cumulants, a joint algorithm is proposed. Firstly, three characteristic parameters is constructed with the utilization of the 4th and 6th order cumulants to identify Multiple Phase Shift Keying (MPSK) and partial Multiple Quadrature Amplitude Modulation (MQAM) modulation types, then the improved constellation subtraction clustering algorithm is combined to separate the remaining modulation patterns, At last, the parameters are integrated to establish a decision tree classifier for unified scheduling. By adopting the method of this article, many signals without prior knowledge are unnecessarily required, and meanwhile the proposed approach maintains the characteristics of simple feature extraction parameters and multiple recognition types. The simulation experiments demonstrate that the associated algorithm is still able to achieve the validity of more than 90%, in the circumstance of the satellite single-mixed signals possessing a Signal-to-Noise Ratio (SNR) of 10 dB.
In order to tackle the problem of single-mixed signal modulation type recognition with low efficiency and poor accuracy in satellite communication, based on clustering characteristics of constellation and high order cumulants, a joint algorithm is proposed. Firstly, three characteristic parameters is constructed with the utilization of the 4th and 6th order cumulants to identify Multiple Phase Shift Keying (MPSK) and partial Multiple Quadrature Amplitude Modulation (MQAM) modulation types, then the improved constellation subtraction clustering algorithm is combined to separate the remaining modulation patterns, At last, the parameters are integrated to establish a decision tree classifier for unified scheduling. By adopting the method of this article, many signals without prior knowledge are unnecessarily required, and meanwhile the proposed approach maintains the characteristics of simple feature extraction parameters and multiple recognition types. The simulation experiments demonstrate that the associated algorithm is still able to achieve the validity of more than 90%, in the circumstance of the satellite single-mixed signals possessing a Signal-to-Noise Ratio (SNR) of 10 dB.
2022, 44(10): 3507-3515.
doi: 10.11999/JEIT210825
Abstract:
Considering the problem that the traditional noise reduction algorithm damages the high Signal-to-Noise Ratio (SNR) signal and reduces the accuracy of signal recognition, a SNR classification algorithm based on convolutional neural network is proposed. The algorithm uses Convolutional Neural Network (CNN) to extract the features of the signal, and uses Fixed K-means (FK-means) algorithm to cluster the extracted features to classify accurately the high and low signal-to-noise ratio signals. The low SNR signal is denoised by the improved median filter algorithm. The improved median filter algorithm adds the correlation mechanism of the front and rear sampling windows on the basis of the traditional median filter to improve the poor effect of the traditional median filter algorithm in dealing with continuous noise. In order to extract the spatial and temporal features of signals fully, a Convolutional neural network and Long-short term memory Parallel (P-CL) network with convolutional neural network and long-short term memory in parallel is proposed. The spatial and temporal features of signals are extracted by convolutional neural network and long-short term memory respectively, and the features are fused and classified. Experiments show that the recognition accuracy of the modulation signal classification model proposed in this paper is 91%, which is 6% higher than that of Convolutional Neural Network and Long-Short Term Memory(CNN-LSTM) network.
Considering the problem that the traditional noise reduction algorithm damages the high Signal-to-Noise Ratio (SNR) signal and reduces the accuracy of signal recognition, a SNR classification algorithm based on convolutional neural network is proposed. The algorithm uses Convolutional Neural Network (CNN) to extract the features of the signal, and uses Fixed K-means (FK-means) algorithm to cluster the extracted features to classify accurately the high and low signal-to-noise ratio signals. The low SNR signal is denoised by the improved median filter algorithm. The improved median filter algorithm adds the correlation mechanism of the front and rear sampling windows on the basis of the traditional median filter to improve the poor effect of the traditional median filter algorithm in dealing with continuous noise. In order to extract the spatial and temporal features of signals fully, a Convolutional neural network and Long-short term memory Parallel (P-CL) network with convolutional neural network and long-short term memory in parallel is proposed. The spatial and temporal features of signals are extracted by convolutional neural network and long-short term memory respectively, and the features are fused and classified. Experiments show that the recognition accuracy of the modulation signal classification model proposed in this paper is 91%, which is 6% higher than that of Convolutional Neural Network and Long-Short Term Memory(CNN-LSTM) network.
2022, 44(10): 3516-3523.
doi: 10.11999/JEIT210830
Abstract:
VLC-WiFi (Visible Light Communication-Wireless Fidelity) heterogeneous networks are becoming a popular short-distance wireless communication solution. However, limited spectrum resources make it difficult for VLC-WiFi heterogeneous network to meet the rapidly growth for data bandwidth of user. Combined with the link transmission performance of the physical layer and the queue buffer delay performance of the media access control layer, an evaluation formula of dynamic link transmission performance and link Quality of Service (QoS) perception level are defined. According to the QoS requirements of data packets, the QoS-aware Cross-Layer Dynamic Resource Allocation (QoS-CLDRA) is proposed. Furthermore, the user matching and power allocation strategy based on non-orthogonal multiple access are designed to improve the system throughput. Simulation results show that the proposed QoS-CLDRA can effectively improve the system throughput and reduce the buffer queue length.
VLC-WiFi (Visible Light Communication-Wireless Fidelity) heterogeneous networks are becoming a popular short-distance wireless communication solution. However, limited spectrum resources make it difficult for VLC-WiFi heterogeneous network to meet the rapidly growth for data bandwidth of user. Combined with the link transmission performance of the physical layer and the queue buffer delay performance of the media access control layer, an evaluation formula of dynamic link transmission performance and link Quality of Service (QoS) perception level are defined. According to the QoS requirements of data packets, the QoS-aware Cross-Layer Dynamic Resource Allocation (QoS-CLDRA) is proposed. Furthermore, the user matching and power allocation strategy based on non-orthogonal multiple access are designed to improve the system throughput. Simulation results show that the proposed QoS-CLDRA can effectively improve the system throughput and reduce the buffer queue length.
2022, 44(10): 3524-3531.
doi: 10.11999/JEIT210802
Abstract:
With the development of 5G mobile communication systems and the optimization of network performance, high-precision and low-complexity path loss prediction models become more important. This paper combined the location of the receiver and transmitter, three-dimensional distance, relative clearance, building density, average height and other environmental characteristics, a machine learning path loss prediction model based on 3D electronic maps is established. And the current 5G hot spot frequency bands data at 700 MHz, 2.4 GHz, 3.5 GHz which measured in large-scale urban scenes are used for training and verification. Results show that the method in this paper has higher prediction accuracy in a complex urban environment, and it is better than the traditional model which is based on the distance between the transmitter and receiver. In addition, a machine learning path loss prediction model based on frequency transfer is also proposed, and the performance is evaluated by using indicators like mean square error, average absolute percentage error, root mean square error, coefficient of determination, etc. The proposed methods can solve the problem of path loss prediction in a complex urban environment with severe building obstruction and without a large amount of test data. Moreover, it can accurately predict the path loss value of the mixed channel consist of line-of-sight and non-line-of-sight in the urban environment.
With the development of 5G mobile communication systems and the optimization of network performance, high-precision and low-complexity path loss prediction models become more important. This paper combined the location of the receiver and transmitter, three-dimensional distance, relative clearance, building density, average height and other environmental characteristics, a machine learning path loss prediction model based on 3D electronic maps is established. And the current 5G hot spot frequency bands data at 700 MHz, 2.4 GHz, 3.5 GHz which measured in large-scale urban scenes are used for training and verification. Results show that the method in this paper has higher prediction accuracy in a complex urban environment, and it is better than the traditional model which is based on the distance between the transmitter and receiver. In addition, a machine learning path loss prediction model based on frequency transfer is also proposed, and the performance is evaluated by using indicators like mean square error, average absolute percentage error, root mean square error, coefficient of determination, etc. The proposed methods can solve the problem of path loss prediction in a complex urban environment with severe building obstruction and without a large amount of test data. Moreover, it can accurately predict the path loss value of the mixed channel consist of line-of-sight and non-line-of-sight in the urban environment.
2022, 44(10): 3532-3540.
doi: 10.11999/JEIT210743
Abstract:
In order to solve the problem of virtual network function migration caused by time-varying network traffic in network slicing, a Virtual Network Function (VNF) migration algorithm based on Federated learning with Bidirectional Gate Recurrent Units (FedBi-GRU) prediction of resource requirements is proposed. Firstly, a VNF migration model of system energy consumption and load balancing is established, and then a framework based on distributed federated learning is introduced to cooperatively train the predictive model. Secondly, considering predicting the resource requirements of VNF, an online training Bidirectional Gate Recurrent Unit (Bi-GRU) algorithm on the basis of the framework is designed. Finally, on the grounds of the resource prediction results, system energy consumption optimization and load balancing are combined, and a Distributed Proximal Policy Optimization (DPPO) migration algorithm is proposed to formulate a VNF migration strategy in advance. The simulation results show that the combination of the two algorithms reduces effectively the energy consumption of the network system and ensures the load balance.
In order to solve the problem of virtual network function migration caused by time-varying network traffic in network slicing, a Virtual Network Function (VNF) migration algorithm based on Federated learning with Bidirectional Gate Recurrent Units (FedBi-GRU) prediction of resource requirements is proposed. Firstly, a VNF migration model of system energy consumption and load balancing is established, and then a framework based on distributed federated learning is introduced to cooperatively train the predictive model. Secondly, considering predicting the resource requirements of VNF, an online training Bidirectional Gate Recurrent Unit (Bi-GRU) algorithm on the basis of the framework is designed. Finally, on the grounds of the resource prediction results, system energy consumption optimization and load balancing are combined, and a Distributed Proximal Policy Optimization (DPPO) migration algorithm is proposed to formulate a VNF migration strategy in advance. The simulation results show that the combination of the two algorithms reduces effectively the energy consumption of the network system and ensures the load balance.
2022, 44(10): 3541-3552.
doi: 10.11999/JEIT210775
Abstract:
In distributed Inverse Synthetic Aperture Radar (ISAR) imaging, if the transmitted waveforms are nonorthogonal, it is difficult to obtain the ideal range image by the traditional matched filtering method, which will affect the azimuth imaging effect. Sparse-based method can replace matched filtering in range profile separation. In this paper, after describing the sparse representation model of range image in a single snapshot, by adjusting the delay of the transmitting and receiving sensors, the range image with multiple receiving sensors can have joint-block sparse characteristics. Then, a Multiple Measurement Vectors Joint Block (MMV-JBlock) algorithm is constructed using Sequential Order One Negative Exponential (SOONE) function to improve the effect of sparse reconstruction. For multiple snapshots, the MMV-JBlock method is used to separate the range image at each snapshot firstly. After aligning the multi-channel range images, the uninterested directional motion and error items in the azimuth phase are compensated. Finally, the sparse method is used to obtain the target azimuth image. The simulation verifies the reconstruction performance of the proposed algorithm under different sparsity and different signal-to-noise ratios, and achieves the imaging of moving targets by distributed ISAR, which validates the effectiveness of the proposed method.
In distributed Inverse Synthetic Aperture Radar (ISAR) imaging, if the transmitted waveforms are nonorthogonal, it is difficult to obtain the ideal range image by the traditional matched filtering method, which will affect the azimuth imaging effect. Sparse-based method can replace matched filtering in range profile separation. In this paper, after describing the sparse representation model of range image in a single snapshot, by adjusting the delay of the transmitting and receiving sensors, the range image with multiple receiving sensors can have joint-block sparse characteristics. Then, a Multiple Measurement Vectors Joint Block (MMV-JBlock) algorithm is constructed using Sequential Order One Negative Exponential (SOONE) function to improve the effect of sparse reconstruction. For multiple snapshots, the MMV-JBlock method is used to separate the range image at each snapshot firstly. After aligning the multi-channel range images, the uninterested directional motion and error items in the azimuth phase are compensated. Finally, the sparse method is used to obtain the target azimuth image. The simulation verifies the reconstruction performance of the proposed algorithm under different sparsity and different signal-to-noise ratios, and achieves the imaging of moving targets by distributed ISAR, which validates the effectiveness of the proposed method.
2022, 44(10): 3553-3565.
doi: 10.11999/JEIT210657
Abstract:
The current satellite ground test system has outstanding real-time attributes, but due to insufficient data mining and analysis, it is difficult to achieve satellite system-level health diagnosis. Comprehensive evaluation needs to be completed manually, and there are problems such as low efficiency and poor versatility. A comprehensive evaluation method for multi-level heterogeneous satellite systems is proposed in this paper. According to the characteristics of the slow, urgent, and key variables in the data, the single-item evaluation generation sheet based on the Gaussian distribution model, the Long Short-Term Memory model (LSTM) and the statistical model is realized respectively. The maximum deviation method is used to realize the combination of subjective and objective weight vector of the analytic hierarchy process and entropy weight method, and comprehensively evaluate the satellite state based on the fuzzy comprehensive evaluation method, and the automation and intelligence of the evaluation process are realized. The system verification is carried out on the small satellite semi-physical simulation platform, and the results show that the evaluation method can effectively evaluate the health status of the satellite system.
The current satellite ground test system has outstanding real-time attributes, but due to insufficient data mining and analysis, it is difficult to achieve satellite system-level health diagnosis. Comprehensive evaluation needs to be completed manually, and there are problems such as low efficiency and poor versatility. A comprehensive evaluation method for multi-level heterogeneous satellite systems is proposed in this paper. According to the characteristics of the slow, urgent, and key variables in the data, the single-item evaluation generation sheet based on the Gaussian distribution model, the Long Short-Term Memory model (LSTM) and the statistical model is realized respectively. The maximum deviation method is used to realize the combination of subjective and objective weight vector of the analytic hierarchy process and entropy weight method, and comprehensively evaluate the satellite state based on the fuzzy comprehensive evaluation method, and the automation and intelligence of the evaluation process are realized. The system verification is carried out on the small satellite semi-physical simulation platform, and the results show that the evaluation method can effectively evaluate the health status of the satellite system.
2022, 44(10): 3566-3573.
doi: 10.11999/JEIT210724
Abstract:
Considering the issue of performance degradation or even failure of the available Inverse Synthetic Aperture Radar (ISAR) object recognition methods based on Deep Convolution Neural Networks (DCNNs) with insufficient training samples, a small- data ISAR object recognition method based on Gaussian Prototypical Network (GPN) is proposed. Firstly, ISAR images are maped into embedding vectors by the embedding network, and then Gaussian prototypes are constructed according to the weighted embedding vectors. Finally, the object category is output according to the Mahalanobis distance from the test samples to all prototypes. Recognition results of the three different types of aircraft show that the proposed method can obtain higher average recognition accuracy under small-data scenarios.
Considering the issue of performance degradation or even failure of the available Inverse Synthetic Aperture Radar (ISAR) object recognition methods based on Deep Convolution Neural Networks (DCNNs) with insufficient training samples, a small- data ISAR object recognition method based on Gaussian Prototypical Network (GPN) is proposed. Firstly, ISAR images are maped into embedding vectors by the embedding network, and then Gaussian prototypes are constructed according to the weighted embedding vectors. Finally, the object category is output according to the Mahalanobis distance from the test samples to all prototypes. Recognition results of the three different types of aircraft show that the proposed method can obtain higher average recognition accuracy under small-data scenarios.
2022, 44(10): 3574-3582.
doi: 10.11999/JEIT210820
Abstract:
Considering the problems of the existing time-frequency analysis of Prolate Spheroidal Wave Functions (PSWFs) signals without explicit expressions, uncontrollable numerical simulation errors, and lack of symmetry in the time-frequency distribution results, Legendre polynomials and Wigner- Ville Distribution (WVD) are introduced in this paper, and an explicit and progressive solution method for PSWFs signal WVD is proposed. According to the error requirements, this method generates the Legendre polynomial WVD self-terms and cross-terms of the required order, and then multiplies them with the corresponding WVD-Legendre coefficients and superimposes linearly them to obtain the explicit and progressive expression of the PSWFs signal WVD. Theoretical and numerical simulation results show that the proposed method can produce an explicit and progressive expression of the PSWFs signal WVD that meets the error requirements, and can effectively maintain the original time-domain and frequency-domain symmetry of the signal. In addition, in the case of the same number of sampling points, compared with the PSWFs signal WVD based on the numerical solution, the PSWFs signal WVD obtained by the proposed method has a higher frequency domain resolution.
Considering the problems of the existing time-frequency analysis of Prolate Spheroidal Wave Functions (PSWFs) signals without explicit expressions, uncontrollable numerical simulation errors, and lack of symmetry in the time-frequency distribution results, Legendre polynomials and Wigner- Ville Distribution (WVD) are introduced in this paper, and an explicit and progressive solution method for PSWFs signal WVD is proposed. According to the error requirements, this method generates the Legendre polynomial WVD self-terms and cross-terms of the required order, and then multiplies them with the corresponding WVD-Legendre coefficients and superimposes linearly them to obtain the explicit and progressive expression of the PSWFs signal WVD. Theoretical and numerical simulation results show that the proposed method can produce an explicit and progressive expression of the PSWFs signal WVD that meets the error requirements, and can effectively maintain the original time-domain and frequency-domain symmetry of the signal. In addition, in the case of the same number of sampling points, compared with the PSWFs signal WVD based on the numerical solution, the PSWFs signal WVD obtained by the proposed method has a higher frequency domain resolution.
2022, 44(10): 3583-3591.
doi: 10.11999/JEIT210696
Abstract:
In order to realize the recognition of human posture in complex and diverse environments, a method based on Frequency Modulated Continuous Wave (FMCW) radar signal is proposed. This method obtains multi-dimensional information of distance, speed and angle by performing 3D fast Fourier transform on the original signal of FMCW radar. After using the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm and the Hampel filter algorithm to solve the noise interference of dynamic or static targets in the range of motion, the convolutional neural network is used to extract the features of multi-dimensional information. Then Low-rank Multimodal Fusion network (LMF) is used to fully fuse the features of multi-dimensional information. Finally, the domain discriminator further obtains domain-independent features, and the activity recognizer obtains the result of gesture recognition. The pre-designed algorithm and the trained network model are carried out on the edge computing platform for experimental verification. Experimental results show that the recognition accuracy of this method can reach 91.5% in complex environments.
In order to realize the recognition of human posture in complex and diverse environments, a method based on Frequency Modulated Continuous Wave (FMCW) radar signal is proposed. This method obtains multi-dimensional information of distance, speed and angle by performing 3D fast Fourier transform on the original signal of FMCW radar. After using the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm and the Hampel filter algorithm to solve the noise interference of dynamic or static targets in the range of motion, the convolutional neural network is used to extract the features of multi-dimensional information. Then Low-rank Multimodal Fusion network (LMF) is used to fully fuse the features of multi-dimensional information. Finally, the domain discriminator further obtains domain-independent features, and the activity recognizer obtains the result of gesture recognition. The pre-designed algorithm and the trained network model are carried out on the edge computing platform for experimental verification. Experimental results show that the recognition accuracy of this method can reach 91.5% in complex environments.
2022, 44(10): 3592-3601.
doi: 10.11999/JEIT210661
Abstract:
To solve the core problem in signal processing of the enhanced LOng RAnge Navigation (eLORAN) system—cycle-identification, a joint algorithm for harsh condition such as high intensity skywave interference and low Signal-Noise-Ratio (SNR) is proposed in this paper. Firstly, based on the improved window function in this method, the characteristic parameters of signal are estimated by spectral division technology, and then the identification of ground and sky wave is realized according to the thought of large number theory. Secondly, in order to reconstruct accurately and remove the skywave while saving the computation, a two-stage adaptive searching and matching algorithm of the characteristic parameters is proposed. Finally, the cycle-identification is realized accurately by the output pseudo-groundwave. The analysis of simulation results show that the proposed algorithm can successfully overcome some disadvantages of the prior art, and realize the recognition and separation of skywave in the environment of low time-delay and high level skywave. In addition, the accuracy rate of cycle-identification is greatly improved combining with the stability of spectral division technology, so as to provide a guarantee for the subsequent demodulation and decoding processes.
To solve the core problem in signal processing of the enhanced LOng RAnge Navigation (eLORAN) system—cycle-identification, a joint algorithm for harsh condition such as high intensity skywave interference and low Signal-Noise-Ratio (SNR) is proposed in this paper. Firstly, based on the improved window function in this method, the characteristic parameters of signal are estimated by spectral division technology, and then the identification of ground and sky wave is realized according to the thought of large number theory. Secondly, in order to reconstruct accurately and remove the skywave while saving the computation, a two-stage adaptive searching and matching algorithm of the characteristic parameters is proposed. Finally, the cycle-identification is realized accurately by the output pseudo-groundwave. The analysis of simulation results show that the proposed algorithm can successfully overcome some disadvantages of the prior art, and realize the recognition and separation of skywave in the environment of low time-delay and high level skywave. In addition, the accuracy rate of cycle-identification is greatly improved combining with the stability of spectral division technology, so as to provide a guarantee for the subsequent demodulation and decoding processes.
2022, 44(10): 3602-3609.
doi: 10.11999/JEIT210763
Abstract:
In view of the limited mapping range of the original Logistic map, the small range of chaotic parameters, and the uneven distribution, a new improved Logistic chaotic map is proposed. The mapping has two parameters\begin{document}$ \mu $\end{document} ![]()
![]()
\begin{document}$ \alpha $\end{document} ![]()
![]()
\begin{document}$ {x_{n - 1}} $\end{document} ![]()
![]()
\begin{document}$ {x_n} $\end{document} ![]()
![]()
In view of the limited mapping range of the original Logistic map, the small range of chaotic parameters, and the uneven distribution, a new improved Logistic chaotic map is proposed. The mapping has two parameters
2022, 44(10): 3610-3617.
doi: 10.11999/JEIT210747
Abstract:
The domestic cryptographic SM9 algorithm is an identity-based cryptographic scheme independently designed by our nation, and has progressively attracted attention from all walks of life. In order to resolve the problem of inefficient verification of the existing Attribute-Based Signature(ABS) schemes, a new attribute-based signature scheme is constructed based on SM9 that supports the dendritic access structure strategy. The signature verification cost of the scheme only requires one bilinear pairing operation and one exponential operation. In addition, the proposed scheme has the function of tracking the identity of the signer, preventing the signer from using anonymity to sign illegally, and avoiding the problem of signature abuse under unconditional anonymity in the traditional attribute-based digital signature scheme. The security analysis results demonstrate that the proposed scheme is unforgeable in random oracle model and can withstand collusion attack. Compared with the existing traceable identity attribute-based signature scheme, the proposed scheme avoids complicated operations for identity tracking algorithm, and has lower signature and verification costs. The experimental results indicate that the computational complexity of the verification has nothing to do with the scale of strategy, and it only takes 2 ms to complete a verification.
The domestic cryptographic SM9 algorithm is an identity-based cryptographic scheme independently designed by our nation, and has progressively attracted attention from all walks of life. In order to resolve the problem of inefficient verification of the existing Attribute-Based Signature(ABS) schemes, a new attribute-based signature scheme is constructed based on SM9 that supports the dendritic access structure strategy. The signature verification cost of the scheme only requires one bilinear pairing operation and one exponential operation. In addition, the proposed scheme has the function of tracking the identity of the signer, preventing the signer from using anonymity to sign illegally, and avoiding the problem of signature abuse under unconditional anonymity in the traditional attribute-based digital signature scheme. The security analysis results demonstrate that the proposed scheme is unforgeable in random oracle model and can withstand collusion attack. Compared with the existing traceable identity attribute-based signature scheme, the proposed scheme avoids complicated operations for identity tracking algorithm, and has lower signature and verification costs. The experimental results indicate that the computational complexity of the verification has nothing to do with the scale of strategy, and it only takes 2 ms to complete a verification.
2022, 44(10): 3618-3626.
doi: 10.11999/JEIT210882
Abstract:
With the establishment of the Intelligent Transportation Systems (ITS), Vehicular Ad-hoc NETworks (VANETs) play great roles in improving traffic safety and efficiency. However, due to the openness and fragility of VANETs, they are vulnerable to various network threats and attacks, and thereby hindering the wide applications of VANETs. To address the requirements for authentication and integrity of transmitted data, identity privacy-preservation, an anonymous online registration and secure authentication protocol is proposed in intelligent VANETs. The protocol enables a vehicle to execute anonymous online registration in transportation systems Trusted Authority (TA) via a public channel. Once validating the real identity, TA can return the private key to the vehicle for subsequent secure authentication via public channel. Thus, the vehicle can generate an authenticated traffic message to a nearby RoadSide Unit (RSU) in real time, so that RSU performs the authentication and integrity verification. This protocol supports anonymous identity traceability, thus TA can revoke the real identity of a malicious vehicle, which has generated some forged messages and caused traffic jams or accidents. In addition, this protocol supports batch authentication and verification of those transmitted traffic messages from different anonymous vehicles. The detailed security analysis and performance evaluation have been conducted. The results demonstrate that the protocol has outstanding advantages on the computational costs of each vehicle and the communication overhead of RSU, and can realize anonymous online registration without establishing secure channel. Therefore, the protocol could be securely and efficiently deployed in intelligent VANETs.
With the establishment of the Intelligent Transportation Systems (ITS), Vehicular Ad-hoc NETworks (VANETs) play great roles in improving traffic safety and efficiency. However, due to the openness and fragility of VANETs, they are vulnerable to various network threats and attacks, and thereby hindering the wide applications of VANETs. To address the requirements for authentication and integrity of transmitted data, identity privacy-preservation, an anonymous online registration and secure authentication protocol is proposed in intelligent VANETs. The protocol enables a vehicle to execute anonymous online registration in transportation systems Trusted Authority (TA) via a public channel. Once validating the real identity, TA can return the private key to the vehicle for subsequent secure authentication via public channel. Thus, the vehicle can generate an authenticated traffic message to a nearby RoadSide Unit (RSU) in real time, so that RSU performs the authentication and integrity verification. This protocol supports anonymous identity traceability, thus TA can revoke the real identity of a malicious vehicle, which has generated some forged messages and caused traffic jams or accidents. In addition, this protocol supports batch authentication and verification of those transmitted traffic messages from different anonymous vehicles. The detailed security analysis and performance evaluation have been conducted. The results demonstrate that the protocol has outstanding advantages on the computational costs of each vehicle and the communication overhead of RSU, and can realize anonymous online registration without establishing secure channel. Therefore, the protocol could be securely and efficiently deployed in intelligent VANETs.
2022, 44(10): 3627-3634.
doi: 10.11999/JEIT210878
Abstract:
In 2019, CAO et al. (doi: 10.11999/JEIT190166) proposed an efficient certificateless aggregate signature scheme which is suitable for multi-party contract signing environment. They demonstrated that their scheme is unforgeable under the random oracle model. However, by the security analysis, it is found that their scheme can not resist public key substitution attacks and coalition attacks of internal signers. In order to solve the above security defects, an improved certificateless aggregate signature scheme is proposed. The new scheme not only satisfies the unforgeability based on the computational Diffie-Hellman problem under the random oracle model, but also resists coalition attacks.
In 2019, CAO et al. (doi: 10.11999/JEIT190166) proposed an efficient certificateless aggregate signature scheme which is suitable for multi-party contract signing environment. They demonstrated that their scheme is unforgeable under the random oracle model. However, by the security analysis, it is found that their scheme can not resist public key substitution attacks and coalition attacks of internal signers. In order to solve the above security defects, an improved certificateless aggregate signature scheme is proposed. The new scheme not only satisfies the unforgeability based on the computational Diffie-Hellman problem under the random oracle model, but also resists coalition attacks.
2022, 44(10): 3635-3642.
doi: 10.11999/JEIT210791
Abstract:
In recent years, new image encryption algorithms have been continuously proposed, but their security has not been fully analyzed and verified. The security of a newly reported image encryption algorithm is analyzed in this paper. The analyzed algorithm achieves the image encryption through pixel scrambling based on variable step length Josephus traversing, pixel substitution based on DNA dynamic encoding, and pixel diffusion in row and column directions. Analysis shows that the secret key design of this algorithm is not practical, and its encryption process also has defects. Under the condition of chosen-plaintext attack, the encryption process of this algorithm is cryptanalyzed, and a corresponding attack algorithm is proposed. Simulation experiments and theoretical analysis confirm the effectiveness and feasibility of the proposed attack algorithm. Finally, in view of issues in the analyzed algorithm and some image encryption algorithms, several suggestions for improvement are given.
In recent years, new image encryption algorithms have been continuously proposed, but their security has not been fully analyzed and verified. The security of a newly reported image encryption algorithm is analyzed in this paper. The analyzed algorithm achieves the image encryption through pixel scrambling based on variable step length Josephus traversing, pixel substitution based on DNA dynamic encoding, and pixel diffusion in row and column directions. Analysis shows that the secret key design of this algorithm is not practical, and its encryption process also has defects. Under the condition of chosen-plaintext attack, the encryption process of this algorithm is cryptanalyzed, and a corresponding attack algorithm is proposed. Simulation experiments and theoretical analysis confirm the effectiveness and feasibility of the proposed attack algorithm. Finally, in view of issues in the analyzed algorithm and some image encryption algorithms, several suggestions for improvement are given.
2022, 44(10): 3643-3649.
doi: 10.11999/JEIT210720
Abstract:
Linear codes play an important role in data storage, information security and secret sharing. Minimal linear codes are the first choice to design secret sharing schemes, so the design of minimal linear codes is one of the important contents of current cryptosystem and coding theory. In this paper, the Walsh spectrum distribution of the selected Boolean functions is studied, and two kinds of minimal linear codes are obtained by using the Walsh spectrum distribution of the functions, then the weight distribution of the codes are determined. The results show that the constructed codes are minimal linear codes that do not satisfy the Ashikhmin-Barg condition, and can be used to design secret sharing schemes with good access structure.
Linear codes play an important role in data storage, information security and secret sharing. Minimal linear codes are the first choice to design secret sharing schemes, so the design of minimal linear codes is one of the important contents of current cryptosystem and coding theory. In this paper, the Walsh spectrum distribution of the selected Boolean functions is studied, and two kinds of minimal linear codes are obtained by using the Walsh spectrum distribution of the functions, then the weight distribution of the codes are determined. The results show that the constructed codes are minimal linear codes that do not satisfy the Ashikhmin-Barg condition, and can be used to design secret sharing schemes with good access structure.
2022, 44(10): 3650-3656.
doi: 10.11999/JEIT210697
Abstract:
Compared with traditional storage, the difficulty of DeoxyriboNucleic Acid (DNA) data storage is that insertion and deletion errors in sequenced reads pose a great challenge to data recovery. For forward error-correcting coded DNA storage with one-base error-correcting capability, a bucket allocation strategy is proposed to improve the decoding accuracy and efficiency. Firstly, all identifiable DNA codes of reads in each cluster are searched and the corresponding valid codes according to the one-base error-correcting capability are determined; Then, for each identifiable DNA code, appropriate coding position (i.e. bucket) according is allocated to its position in a read; Finally, the consensus code for each bucket is determined using majority voting strategy. Simulation results show that the proposed method can correct more than 94% errors at the coverage of 20X when error rate is 5% or 10%, and correct more than 90% errors at the coverage of 60X when error rate is 15%.
Compared with traditional storage, the difficulty of DeoxyriboNucleic Acid (DNA) data storage is that insertion and deletion errors in sequenced reads pose a great challenge to data recovery. For forward error-correcting coded DNA storage with one-base error-correcting capability, a bucket allocation strategy is proposed to improve the decoding accuracy and efficiency. Firstly, all identifiable DNA codes of reads in each cluster are searched and the corresponding valid codes according to the one-base error-correcting capability are determined; Then, for each identifiable DNA code, appropriate coding position (i.e. bucket) according is allocated to its position in a read; Finally, the consensus code for each bucket is determined using majority voting strategy. Simulation results show that the proposed method can correct more than 94% errors at the coverage of 20X when error rate is 5% or 10%, and correct more than 90% errors at the coverage of 60X when error rate is 15%.
2022, 44(10): 3657-3665.
doi: 10.11999/JEIT210677
Abstract:
Memristor is an ideal device to realize artificial synapses due to its low power consumption, memory ability and nanometer size. In order to construct a simple, efficient and comprehensive associative memory circuit, a simple neuron circuit and a synaptic circuit based on the voltage-controlled threshold memristor is proposed. Then according to Pavlov’s associative memory model, the corresponding associative memory circuit is designed. This circuit has a simple structure and only contains three neurons and a memristive synapse, which can effectively reduce the network complexity and power consumption. what’s more, the circuit can simulate the full function of associative memory behavior. It not only realizes functions of learning, forgetting, accelerated-learning, deceleration-forgetting and deceleration-nature-forgetting, but also the learning rate and natural forgetting rate can be automatically adjusted according to the number of learning, so that the circuit is more bionic. In addition, the designed circuit agrees well with the Ebbinghaus forgetting curve, which enlarges the scope of application of the circuit.
Memristor is an ideal device to realize artificial synapses due to its low power consumption, memory ability and nanometer size. In order to construct a simple, efficient and comprehensive associative memory circuit, a simple neuron circuit and a synaptic circuit based on the voltage-controlled threshold memristor is proposed. Then according to Pavlov’s associative memory model, the corresponding associative memory circuit is designed. This circuit has a simple structure and only contains three neurons and a memristive synapse, which can effectively reduce the network complexity and power consumption. what’s more, the circuit can simulate the full function of associative memory behavior. It not only realizes functions of learning, forgetting, accelerated-learning, deceleration-forgetting and deceleration-nature-forgetting, but also the learning rate and natural forgetting rate can be automatically adjusted according to the number of learning, so that the circuit is more bionic. In addition, the designed circuit agrees well with the Ebbinghaus forgetting curve, which enlarges the scope of application of the circuit.
2022, 44(10): 3666-3672.
doi: 10.11999/JEIT210781
Abstract:
A compact power divider with ultra-wide stopband for harmonic suppression based on resonator slow-wave transmission lines is proposed in this paper. The resonator slow-wave transmission line is consisted of rectangular resonators, a T-type resonator and serpentine lines, and it is used to replace the conventional quarter wavelength transmission line of the power divider. The proposed power divider is only 37.4% size of the conventional microstrip power divider. The experimental results show that the bandwidth of return loss greater than 10 dB is 0.1 GHz～1.19 GHz. In the range from 2.2 GHz to 11.05 GHz, the attenuation is more than 20 dB, so the power divider has an ultra-wide stopband with harmonic suppression performance. The simulations agree well with the measured results, verifiy the feasibility of the proposed design method.
A compact power divider with ultra-wide stopband for harmonic suppression based on resonator slow-wave transmission lines is proposed in this paper. The resonator slow-wave transmission line is consisted of rectangular resonators, a T-type resonator and serpentine lines, and it is used to replace the conventional quarter wavelength transmission line of the power divider. The proposed power divider is only 37.4% size of the conventional microstrip power divider. The experimental results show that the bandwidth of return loss greater than 10 dB is 0.1 GHz～1.19 GHz. In the range from 2.2 GHz to 11.05 GHz, the attenuation is more than 20 dB, so the power divider has an ultra-wide stopband with harmonic suppression performance. The simulations agree well with the measured results, verifiy the feasibility of the proposed design method.
2022, 44(10): 3673-3682.
doi: 10.11999/JEIT210795
Abstract:
Considering the problem that the intermittent fault signal of the electronic system is greatly affected by noise and has a lot of redundant information, which results in the limitation of the deep neural network model to evaluate the severity of the intermittent fault. A method for evaluating the severity of intermittent faults based on Variational Mode Decomposition-Gated Recurrent Units (VMD-GRU) is proposed. Firstly, all Intrinsic Mode Function (IMF) components are adaptively decomposed on intermittent fault signals through Variational Mode Decomposition (VMD). Then the sensitivity analysis of the IMF components is performed to select the sensitive components, and the differential enhanced energy operator is used to construct the severity sensitivity factor. Finally, the severity sensitivity factor is used to train the Gated Recurrent Units (GRU) recurrent neural network severity evaluation model. Through the evaluation of intermittent faults of different severity injected into the key circuits of electronic systems, the results show that this method has a strong ability to evaluate the severity of intermittent faults, and is more accurate and effective in evaluating the severity of intermittent faults.
Considering the problem that the intermittent fault signal of the electronic system is greatly affected by noise and has a lot of redundant information, which results in the limitation of the deep neural network model to evaluate the severity of the intermittent fault. A method for evaluating the severity of intermittent faults based on Variational Mode Decomposition-Gated Recurrent Units (VMD-GRU) is proposed. Firstly, all Intrinsic Mode Function (IMF) components are adaptively decomposed on intermittent fault signals through Variational Mode Decomposition (VMD). Then the sensitivity analysis of the IMF components is performed to select the sensitive components, and the differential enhanced energy operator is used to construct the severity sensitivity factor. Finally, the severity sensitivity factor is used to train the Gated Recurrent Units (GRU) recurrent neural network severity evaluation model. Through the evaluation of intermittent faults of different severity injected into the key circuits of electronic systems, the results show that this method has a strong ability to evaluate the severity of intermittent faults, and is more accurate and effective in evaluating the severity of intermittent faults.
2022, 44(10): 3683-3696.
doi: 10.11999/JEIT220625
Abstract:
As a kind of low-complexity and non-coherent information transmission schemes, the differential chaotic communication system has been widely studied because of its good performance against multipath fading. Recently, a series of fruitful researches on Differential Chaos Shift Keying (DCSK) have been carried out, and the signal design and performance optimization for differential chaotic communications have also been developed. Therefore, the main research progresses in signal design of differential chaotic communications are surveyed in detail in the paper from the following four perspectives: design of signal frames, design of orthogonal multilevel signals, design of constellation diagrams and design of multicarrier signals. In addition, the research works on noise suppression aided performance optimization, index modulation aided performance optimization and chaotic shape forming aided performance optimization for differential chaotic communications are summarized in the paper.
As a kind of low-complexity and non-coherent information transmission schemes, the differential chaotic communication system has been widely studied because of its good performance against multipath fading. Recently, a series of fruitful researches on Differential Chaos Shift Keying (DCSK) have been carried out, and the signal design and performance optimization for differential chaotic communications have also been developed. Therefore, the main research progresses in signal design of differential chaotic communications are surveyed in detail in the paper from the following four perspectives: design of signal frames, design of orthogonal multilevel signals, design of constellation diagrams and design of multicarrier signals. In addition, the research works on noise suppression aided performance optimization, index modulation aided performance optimization and chaotic shape forming aided performance optimization for differential chaotic communications are summarized in the paper.
2022, 44(10): 3697-3708.
doi: 10.11999/JEIT210790
Abstract:
Object detection is one of the basic tasks and research hotspots in the field of computer vision. The YOLO (You Only Look Once) frames object detection is a regression problem to implement end-to-end training and detection. YOLO becomes the leading object detector due to its good speed-accuracy balance, which has been successfully studied, improved, and applied to many different fields. YOLO series and its important improvements and applications are investigated in detail. Firstly, the YOLO family and important improvements are systematically summarized, including YOLOv1-v4, YOLOv5, Scaled-YOLOv4, YOLOR, and the latest YOLOX. Then, important backbone and loss functions in YOLO are analyzed and summarized in detail. Next, the application of YOLO is systematically classified and summarized according to different improvement ideas or scenarios, such as attention mechanisms, three-dimensional scenes, aerial scenes, edge computing, etc. Finally, the characteristics of the YOLO series are summarized and the possible improvement ideas and research trends are analyzed in combination with the latest literature.
Object detection is one of the basic tasks and research hotspots in the field of computer vision. The YOLO (You Only Look Once) frames object detection is a regression problem to implement end-to-end training and detection. YOLO becomes the leading object detector due to its good speed-accuracy balance, which has been successfully studied, improved, and applied to many different fields. YOLO series and its important improvements and applications are investigated in detail. Firstly, the YOLO family and important improvements are systematically summarized, including YOLOv1-v4, YOLOv5, Scaled-YOLOv4, YOLOR, and the latest YOLOX. Then, important backbone and loss functions in YOLO are analyzed and summarized in detail. Next, the application of YOLO is systematically classified and summarized according to different improvement ideas or scenarios, such as attention mechanisms, three-dimensional scenes, aerial scenes, edge computing, etc. Finally, the characteristics of the YOLO series are summarized and the possible improvement ideas and research trends are analyzed in combination with the latest literature.
