Email alert
2023 Vol. 45, No. 10
Display Method:
2023, 45(10): 3439-3457.
doi: 10.11999/JEIT221051
Abstract:
Sign Language Recognition (SLR) technology is an important technical means to break the communication barrier between hearing-impaired people and healthy people. The sign language datasets, evaluation indicators and sign language recognition methods in recent years are summarized. Firstly, the sign language dataset is systematically summarized and the development trend of the dataset of sign language recognition methods is analyzed. Secondly, the evaluation indicator of sign language recognition method is introduced in detail. Then, according to the content of sign language expression and the features used in sign language recognition methods, isolated word sign language recognition methods and continuous sign language recognition methods, sign language recognition methods relying only on hand features and sign language recognition methods of multi feature fusion are summarized and analyzed. Finally, the challenges and development direction of sign language recognition technology are discussed.
Sign Language Recognition (SLR) technology is an important technical means to break the communication barrier between hearing-impaired people and healthy people. The sign language datasets, evaluation indicators and sign language recognition methods in recent years are summarized. Firstly, the sign language dataset is systematically summarized and the development trend of the dataset of sign language recognition methods is analyzed. Secondly, the evaluation indicator of sign language recognition method is introduced in detail. Then, according to the content of sign language expression and the features used in sign language recognition methods, isolated word sign language recognition methods and continuous sign language recognition methods, sign language recognition methods relying only on hand features and sign language recognition methods of multi feature fusion are summarized and analyzed. Finally, the challenges and development direction of sign language recognition technology are discussed.
2023, 45(10): 3458-3467.
doi: 10.11999/JEIT221449
Abstract:
Movement intention based Brain-Computer Interfaces (BCIs) have important research significance and application value in motor enhancement, replacement and rehabilitation. Among them, Motor Imagery (MI) is the most commonly used BCI paradigm to represent motor intention. However, traditional MI-BCIs usually focus on the recognition of the intention of different limbs, and the classification accuracies are relatively low, which restricts fine motor control and rehabilitation. To solve the above problems, in recent years, researchers have carried out a series of meaningful explorations in coding and decoding of scalp ElectroEncephaloGram (EEG) from three aspects: specific parts of a single limb movement intention, kinematic and kinetics intention, and mismatch between movement and expectation. On the basis of the above research, some typical applications to high freedom motor control and stroke rehabilitation have been developed. The research progress in this field from the related paradigms of scalp EEG coding and decoding of motor intention and its BCI application is reviewed. Besides, the existing challenges and possible solutions are discussed, considering to promote the in-depth research and application of motor intention based BCIs.
Movement intention based Brain-Computer Interfaces (BCIs) have important research significance and application value in motor enhancement, replacement and rehabilitation. Among them, Motor Imagery (MI) is the most commonly used BCI paradigm to represent motor intention. However, traditional MI-BCIs usually focus on the recognition of the intention of different limbs, and the classification accuracies are relatively low, which restricts fine motor control and rehabilitation. To solve the above problems, in recent years, researchers have carried out a series of meaningful explorations in coding and decoding of scalp ElectroEncephaloGram (EEG) from three aspects: specific parts of a single limb movement intention, kinematic and kinetics intention, and mismatch between movement and expectation. On the basis of the above research, some typical applications to high freedom motor control and stroke rehabilitation have been developed. The research progress in this field from the related paradigms of scalp EEG coding and decoding of motor intention and its BCI application is reviewed. Besides, the existing challenges and possible solutions are discussed, considering to promote the in-depth research and application of motor intention based BCIs.
2023, 45(10): 3468-3482.
doi: 10.11999/JEIT221402
Abstract:
Underwater image object detection is one of the core technologies of underwater intelligent exploration, which is widely used in industrial and military fields. The breakthrough of deep learning related technologies has brought new opportunities for the development of underwater image object detection, but the current reviews are relatively old and lack a certain degree of systematicness and comprehensiveness. In this paper, the research of underwater visible and sonar image detection based on deep learning is summarized and analyzed in detail. Firstly, the general object detection algorithm framework based on deep learning is sorted out, including six elements: backbone, neck, head, training algorithm, inference strategy, and evaluation criteria, and the problems of each element and the latest research work are systematically summarized; Then, the latest progresses of underwater visible image object detection are investigated and summarized from three aspects: data set, model design, and training method; Meanwhile, the works related to underwater sonar image detection are summarized and analyzed, including forward-looking sonar, side-scanning sonar and synthetic aperture sonar. Finally, the research trend of underwater image object detection is discussed based on the latest research on deep learning.
Underwater image object detection is one of the core technologies of underwater intelligent exploration, which is widely used in industrial and military fields. The breakthrough of deep learning related technologies has brought new opportunities for the development of underwater image object detection, but the current reviews are relatively old and lack a certain degree of systematicness and comprehensiveness. In this paper, the research of underwater visible and sonar image detection based on deep learning is summarized and analyzed in detail. Firstly, the general object detection algorithm framework based on deep learning is sorted out, including six elements: backbone, neck, head, training algorithm, inference strategy, and evaluation criteria, and the problems of each element and the latest research work are systematically summarized; Then, the latest progresses of underwater visible image object detection are investigated and summarized from three aspects: data set, model design, and training method; Meanwhile, the works related to underwater sonar image detection are summarized and analyzed, including forward-looking sonar, side-scanning sonar and synthetic aperture sonar. Finally, the research trend of underwater image object detection is discussed based on the latest research on deep learning.
2023, 45(10): 3483-3491.
doi: 10.11999/JEIT221244
Abstract:
Learning-based Multiple-View Stereo (MVS) aims to reconstruct dense 3D scene representation. However, previous methods utilize additional 2D network modules to learn the cross view visibility for cost aggregation, ignoring the consistency assumption of 2D contextual features in the 3D depth direction. In addition, the current multi-stage depth inference model still requires a high depth sampling rate, and depth hypothesis is sampled within static and preset depth range, which is prone to generate errorneous depth inference in the object boundary and occluded area. To alleviate these problems, a multi-view stereo network based on edge assisted epipolar Transformer is proposed. The improvements of this work over the state of the art are as: Depth regression is replaced by the multi-depth hypotheses classification to ensure the accuracy with limited depth sampling rate and GPU consumption. Epipolar Transformer block is developed for reliable cross view cost aggregation, and edge detection branch is designed to constrain the consistency of edge features in the epipolar direction. A dynamic depth range sampling mechanism based on probabilistic cost volume is applied to improve the accuracy of uncertain areas. Comprehensive comparisons with the state of the art are conducted on public benchmarks, which indicate that the proposed method can reconstruct dense scene representations with limited memory bottleblock. Specifically, compared with Cas-MVSNet, the memory consumption is reducted by 35%, the depth sampling rate is reduced by about 50%, and the overall error on DTU datasets is reduced from 0.355 to 0.325.
Learning-based Multiple-View Stereo (MVS) aims to reconstruct dense 3D scene representation. However, previous methods utilize additional 2D network modules to learn the cross view visibility for cost aggregation, ignoring the consistency assumption of 2D contextual features in the 3D depth direction. In addition, the current multi-stage depth inference model still requires a high depth sampling rate, and depth hypothesis is sampled within static and preset depth range, which is prone to generate errorneous depth inference in the object boundary and occluded area. To alleviate these problems, a multi-view stereo network based on edge assisted epipolar Transformer is proposed. The improvements of this work over the state of the art are as: Depth regression is replaced by the multi-depth hypotheses classification to ensure the accuracy with limited depth sampling rate and GPU consumption. Epipolar Transformer block is developed for reliable cross view cost aggregation, and edge detection branch is designed to constrain the consistency of edge features in the epipolar direction. A dynamic depth range sampling mechanism based on probabilistic cost volume is applied to improve the accuracy of uncertain areas. Comprehensive comparisons with the state of the art are conducted on public benchmarks, which indicate that the proposed method can reconstruct dense scene representations with limited memory bottleblock. Specifically, compared with Cas-MVSNet, the memory consumption is reducted by 35%, the depth sampling rate is reduced by about 50%, and the overall error on DTU datasets is reduced from 0.355 to 0.325.
2023, 45(10): 3492-3501.
doi: 10.11999/JEIT220960
Abstract:
Principal Skewness Analysis (PSA), as a third-order extension of Principal Component Analysis (PCA), is often used for blind image separation, SAR image denoising, and hyperspectral feature extraction. However, the existing PSA algorithm can only obtain approximate solutions, which will affect the accuracy of subsequent image processing. In view of this problem, a high-precision Parallel Principal Skewness Analysis (PPSA) algorithm based on the existing PSA algorithm is proposed. The PPSA algorithm considers fully the data structure, and selects the eigenvectors of all slices of the co-skewness tensor as the initial value of the iteration, which can accurately obtain the actual solution. Simulation experiments and actual remote sensing image experiments verify the effectiveness and superiority of the PSA algorithm.
Principal Skewness Analysis (PSA), as a third-order extension of Principal Component Analysis (PCA), is often used for blind image separation, SAR image denoising, and hyperspectral feature extraction. However, the existing PSA algorithm can only obtain approximate solutions, which will affect the accuracy of subsequent image processing. In view of this problem, a high-precision Parallel Principal Skewness Analysis (PPSA) algorithm based on the existing PSA algorithm is proposed. The PPSA algorithm considers fully the data structure, and selects the eigenvectors of all slices of the co-skewness tensor as the initial value of the iteration, which can accurately obtain the actual solution. Simulation experiments and actual remote sensing image experiments verify the effectiveness and superiority of the PSA algorithm.
2023, 45(10): 3502-3510.
doi: 10.11999/JEIT221041
Abstract:
Parkinson’s disease is a common chronic neurological disease, and dysarthria is one of the early symptoms of this disease. The auxiliary diagnosis and treatment of Parkinson’s disease based on speech is helpful for early detection and observation of the development of this disease. Traditional methods evaluate often Parkinson’s disease by calculating the parameters of speech features (such as Jitter, Shimmer, etc.). However, these features may not fully reflect all pathological phenomena, which affects the accuracy of detection and evaluation. In order to extract better the pathological information from speech of patients with Parkinson’s disease and improve the accuracy of detection and evaluation, a Parkinson’s disease detection method based on masking self-supervised speech feature extraction is proposed. First, Mel spectrogram features are extracted from the original speech of Parkinson’s disease patients, and the global temporal representation with rich pathological features is obtained. Then, partial Mel spectrogram features are masked, and the masked parts are reconstructed by masking self-supervised model, so as to learn a higher-level representation of speech features of Parkinson’s disease patients. In order to solve the problem of the scarcity of Parkinson’s disease speech data, the masking self-supervised model will first be pre-trained on LibriSpeech public data set, and then based on the idea of transfer learning, the pre-trained model will be fine-tuned and weighted summed on Parkinson’s disease speech data. Thus, the feature representation learning performance of the proposed masking self-supervised model can be improved. Finally, random forest classifier and support vector machine classifier are used to classify the extracted speech features to achieve the detection of Parkinson’s disease. The effectiveness of the masking self-supervised model is verified on MaxLittle public data set and our self-collected data set by ten-fold cross-validation. The results show that, compared with the traditional Mel spectrogram feature detection method and other classical self-supervised feature extraction methods, the proposed method has significantly improved the Accuracy, True Positive Rate and True Negative Rate performance.
Parkinson’s disease is a common chronic neurological disease, and dysarthria is one of the early symptoms of this disease. The auxiliary diagnosis and treatment of Parkinson’s disease based on speech is helpful for early detection and observation of the development of this disease. Traditional methods evaluate often Parkinson’s disease by calculating the parameters of speech features (such as Jitter, Shimmer, etc.). However, these features may not fully reflect all pathological phenomena, which affects the accuracy of detection and evaluation. In order to extract better the pathological information from speech of patients with Parkinson’s disease and improve the accuracy of detection and evaluation, a Parkinson’s disease detection method based on masking self-supervised speech feature extraction is proposed. First, Mel spectrogram features are extracted from the original speech of Parkinson’s disease patients, and the global temporal representation with rich pathological features is obtained. Then, partial Mel spectrogram features are masked, and the masked parts are reconstructed by masking self-supervised model, so as to learn a higher-level representation of speech features of Parkinson’s disease patients. In order to solve the problem of the scarcity of Parkinson’s disease speech data, the masking self-supervised model will first be pre-trained on LibriSpeech public data set, and then based on the idea of transfer learning, the pre-trained model will be fine-tuned and weighted summed on Parkinson’s disease speech data. Thus, the feature representation learning performance of the proposed masking self-supervised model can be improved. Finally, random forest classifier and support vector machine classifier are used to classify the extracted speech features to achieve the detection of Parkinson’s disease. The effectiveness of the masking self-supervised model is verified on MaxLittle public data set and our self-collected data set by ten-fold cross-validation. The results show that, compared with the traditional Mel spectrogram feature detection method and other classical self-supervised feature extraction methods, the proposed method has significantly improved the Accuracy, True Positive Rate and True Negative Rate performance.
2023, 45(10): 3511-3519.
doi: 10.11999/JEIT221034
Abstract:
Tunable Diode Laser Absorption Tomography (TDLAT) is an important non-intrusive combustion diagnostic technology, which can be used to reconstruct two-dimensional cross-sectional distributions of flow-field parameters such as gas temperature and concentration in the combustion field. In this paper, Vision Transformer (ViT) and multi-scale features merging are introduced into TDLAT to study the nonlinear mapping between a limited number of measurement data and the temperature distribution in the entire tomographic filed. Temperature tomography network (HVTMFnet) is proposed based on the hierarchical Vision Transformer (ViT) and Multi-scale Features merging. By extracting and merging the local and global correlation characteristics of TDLAT measurement data, HVTMFnet reconstructs the hierarchical temperature distribution in the entire tomographic field. Both simulations and lab-scale experiments with TDLAT system show that HVTMFnet retrieves better-quality temperature images than existing temperature tomography schemes based on Convolutional Neural Network (CNN) and residual network. In comparison to the temperature tomography scheme based on CNN, HVTMFnet reduces the reconstruction error by 49.2%～72.1%.
Tunable Diode Laser Absorption Tomography (TDLAT) is an important non-intrusive combustion diagnostic technology, which can be used to reconstruct two-dimensional cross-sectional distributions of flow-field parameters such as gas temperature and concentration in the combustion field. In this paper, Vision Transformer (ViT) and multi-scale features merging are introduced into TDLAT to study the nonlinear mapping between a limited number of measurement data and the temperature distribution in the entire tomographic filed. Temperature tomography network (HVTMFnet) is proposed based on the hierarchical Vision Transformer (ViT) and Multi-scale Features merging. By extracting and merging the local and global correlation characteristics of TDLAT measurement data, HVTMFnet reconstructs the hierarchical temperature distribution in the entire tomographic field. Both simulations and lab-scale experiments with TDLAT system show that HVTMFnet retrieves better-quality temperature images than existing temperature tomography schemes based on Convolutional Neural Network (CNN) and residual network. In comparison to the temperature tomography scheme based on CNN, HVTMFnet reduces the reconstruction error by 49.2%～72.1%.
2023, 45(10): 3520-3528.
doi: 10.11999/JEIT221180
Abstract:
COrona VIrus Disease 2019(COVID-19) is a serious threat to human health, deep learning computer aided diagnosis method can effectively improve the diagnosis efficiency. But deep learning models have usually complex structure which have large number of parameters and computations, therefore, a Dense Reparameter Lightweight Transformer(DRLTransformer) for COVID-19 CT recognition is proposed. Firstly, reparameter dense block and hierarchical Transformer are proposed to improve lightweight degree of model, which can improve computation speed and reduce parameters without losing model performance. Secondly, in order to fully extract global and local information of lesions, using hierarchical Transformer enhance global attention on local feature relevance, which use grouping to extract global features and fused between different groups to obtain multi-level information, and then information fusion is used to further improve interaction of intra group and inter group features. In addition, all global features are aggregated to achieve deep fusion of deep and shallow features. Finally, comparative experiments in COVID-19 CT dataset, the results show that the parameters and computations of DRLTransformer are 1.47 M and 81.232 M. Compared to Dense Convolutional Network(DenseNet), parameters are reduced by 29 times and computations are reduced by 23 times. The model proposed in this paper has positive implications for computer aided diagnosis of COVID-19 and provides a new idea for lightweight deep learning model.
COrona VIrus Disease 2019(COVID-19) is a serious threat to human health, deep learning computer aided diagnosis method can effectively improve the diagnosis efficiency. But deep learning models have usually complex structure which have large number of parameters and computations, therefore, a Dense Reparameter Lightweight Transformer(DRLTransformer) for COVID-19 CT recognition is proposed. Firstly, reparameter dense block and hierarchical Transformer are proposed to improve lightweight degree of model, which can improve computation speed and reduce parameters without losing model performance. Secondly, in order to fully extract global and local information of lesions, using hierarchical Transformer enhance global attention on local feature relevance, which use grouping to extract global features and fused between different groups to obtain multi-level information, and then information fusion is used to further improve interaction of intra group and inter group features. In addition, all global features are aggregated to achieve deep fusion of deep and shallow features. Finally, comparative experiments in COVID-19 CT dataset, the results show that the parameters and computations of DRLTransformer are 1.47 M and 81.232 M. Compared to Dense Convolutional Network(DenseNet), parameters are reduced by 29 times and computations are reduced by 23 times. The model proposed in this paper has positive implications for computer aided diagnosis of COVID-19 and provides a new idea for lightweight deep learning model.
2023, 45(10): 3529-3537.
doi: 10.11999/JEIT221204
Abstract:
Multi-modal medical images can effectively fuse anatomical images and functional images. It reflects the functional and anatomical information within the body on the same image, which gives rise to critical clinical implications. How to utilize efficiently the comprehensive representation capabilities of multimodal medical image information and how to extract adequately cross-scale contextual information are key questions. In this paper, a Transformer segmentation model for PET/CT images with cross-modal, cross-scale and cross-dimension is proposed. The main improvements of the model are as follows: Firstly, PET/CT backbone branch, PET auxiliary branch, and CT auxiliary branch are designed to extract multi-modal image information in the encoder section; Secondly, a cross-modal and cross-dimensional attention module is designed in the skip connection part. The valid information in each dimension of the cross-modal images is captured by this module from both modal and dimensional views; Thirdly, a cross-scale Transformer module is designed at the bottleneck level. Deep semantic information and shallow spatial information are adaptively fused by this model, which can enable the network to learn more contextual information and obtain cross-scale global information; Finally, a multi-scale adaptive decoding feature fusion module is proposed in the decoder part. The multi-scale feature maps with different levels of detail are aggregated and fully utilized in the decoding path, and the noise introduced by upsampling is mitigated in this module. The effectiveness of the algorithm is verified by using a clinical multi-modal lung medical image dataset. All results show that the Acc, Recall, Dice, Voe, Rvd and Miou of the proposed model for lung lesion segmentation are: 97.99%, 94.29%, 95.32%, 92.74%, 92.95% and 90.14%. For the segmentation of lung lesions with complex shapes, it has high accuracy and relatively low redundancy.
Multi-modal medical images can effectively fuse anatomical images and functional images. It reflects the functional and anatomical information within the body on the same image, which gives rise to critical clinical implications. How to utilize efficiently the comprehensive representation capabilities of multimodal medical image information and how to extract adequately cross-scale contextual information are key questions. In this paper, a Transformer segmentation model for PET/CT images with cross-modal, cross-scale and cross-dimension is proposed. The main improvements of the model are as follows: Firstly, PET/CT backbone branch, PET auxiliary branch, and CT auxiliary branch are designed to extract multi-modal image information in the encoder section; Secondly, a cross-modal and cross-dimensional attention module is designed in the skip connection part. The valid information in each dimension of the cross-modal images is captured by this module from both modal and dimensional views; Thirdly, a cross-scale Transformer module is designed at the bottleneck level. Deep semantic information and shallow spatial information are adaptively fused by this model, which can enable the network to learn more contextual information and obtain cross-scale global information; Finally, a multi-scale adaptive decoding feature fusion module is proposed in the decoder part. The multi-scale feature maps with different levels of detail are aggregated and fully utilized in the decoding path, and the noise introduced by upsampling is mitigated in this module. The effectiveness of the algorithm is verified by using a clinical multi-modal lung medical image dataset. All results show that the Acc, Recall, Dice, Voe, Rvd and Miou of the proposed model for lung lesion segmentation are: 97.99%, 94.29%, 95.32%, 92.74%, 92.95% and 90.14%. For the segmentation of lung lesions with complex shapes, it has high accuracy and relatively low redundancy.
2023, 45(10): 3538-3546.
doi: 10.11999/JEIT221500
Abstract:
Most of the current audiovisual separation models are mostly based on simple splicing of video features and audio features, without fully considering the interrelationship of each modality, resulting in the underutilization of visual information, a new model is proposed to address this issue. Hence, in this paper, the interrelationship of each modality is taken into consideration. In addition, a multi-headed attention mechanism is used to combine the Farneback algorithm and the U-Net network to propose a cross-modal fusion optical Flow-Audio Visual Speech Separation (Flow-AVSS) model. The motion features and lip features are respectively extracted by the Farneback algorithm and the lightweight network ShuffleNet v2. Furthermore, the motion features are affine transformed with the lip features, and the visual features are obtained by the Temporal CoNvolution module (TCN). In order to utilize sufficiently the visual information, the multi-headed attention mechanism is used in the feature fusion to fuse the visual features with the audio features across modalities. Finally, the fused audio-visual features are passed through the U-Net separation network to obtain the separated speech. Using Perceptual Evaluation of Speech Quality (PESQ), Short-Time Objective Intelligibility (STOI), and Source-to-Distortion Ratio (SDR) evaluation metrics, experimental tests are conducted on the AVspeech dataset. It is shown that the performance of the proposed method is improved by 2.23 dB and 1.68 dB compared with the pure speech separation network or the audio-visual separation network based on feature splicing. Thus, it is indicated that the feature fusion based on the cross-modal attention can make fuller use of the individual modal correlations. Besides, the increased lip motion features can effectively improve the robustness of video features and improve the separation effect.
Most of the current audiovisual separation models are mostly based on simple splicing of video features and audio features, without fully considering the interrelationship of each modality, resulting in the underutilization of visual information, a new model is proposed to address this issue. Hence, in this paper, the interrelationship of each modality is taken into consideration. In addition, a multi-headed attention mechanism is used to combine the Farneback algorithm and the U-Net network to propose a cross-modal fusion optical Flow-Audio Visual Speech Separation (Flow-AVSS) model. The motion features and lip features are respectively extracted by the Farneback algorithm and the lightweight network ShuffleNet v2. Furthermore, the motion features are affine transformed with the lip features, and the visual features are obtained by the Temporal CoNvolution module (TCN). In order to utilize sufficiently the visual information, the multi-headed attention mechanism is used in the feature fusion to fuse the visual features with the audio features across modalities. Finally, the fused audio-visual features are passed through the U-Net separation network to obtain the separated speech. Using Perceptual Evaluation of Speech Quality (PESQ), Short-Time Objective Intelligibility (STOI), and Source-to-Distortion Ratio (SDR) evaluation metrics, experimental tests are conducted on the AVspeech dataset. It is shown that the performance of the proposed method is improved by 2.23 dB and 1.68 dB compared with the pure speech separation network or the audio-visual separation network based on feature splicing. Thus, it is indicated that the feature fusion based on the cross-modal attention can make fuller use of the individual modal correlations. Besides, the increased lip motion features can effectively improve the robustness of video features and improve the separation effect.
2023, 45(10): 3547-3557.
doi: 10.11999/JEIT220973
Abstract:
This paper proposes an Energy Harvesting (EH)-assisted mining IoT intelligent computation offloading method for the mine IoT device with limited computing, energy, and memory resources and smart mining scenario with a large number of latency-sensitive computation tasks. Mobile Edge Computing (MEC) technology is used to assist task computing of mine IoT devices, and EH technology is investigated to power energy-constrained mine IoT devices. The intelligent computation offloading mechanism based on Q-learning can dynamically explore and optimize computation offloading policy under the condition of an unknown precise mine system model. In addition, a computation offloading mechanism based on Deep Deterministic Policy Gradient (DDPG) is proposed. The curse of dimensionality in complex mining scenarios is resolved, the discretization error caused by policy discretization is reduced, and the computation offloading performance is further improved. Theoretical analysis and simulation results verify that the proposed mechanism can reduce energy consumption, computing delays, and task failure rate. This helps ensure safety and improve the production efficiency of IoT in mining.
This paper proposes an Energy Harvesting (EH)-assisted mining IoT intelligent computation offloading method for the mine IoT device with limited computing, energy, and memory resources and smart mining scenario with a large number of latency-sensitive computation tasks. Mobile Edge Computing (MEC) technology is used to assist task computing of mine IoT devices, and EH technology is investigated to power energy-constrained mine IoT devices. The intelligent computation offloading mechanism based on Q-learning can dynamically explore and optimize computation offloading policy under the condition of an unknown precise mine system model. In addition, a computation offloading mechanism based on Deep Deterministic Policy Gradient (DDPG) is proposed. The curse of dimensionality in complex mining scenarios is resolved, the discretization error caused by policy discretization is reduced, and the computation offloading performance is further improved. Theoretical analysis and simulation results verify that the proposed mechanism can reduce energy consumption, computing delays, and task failure rate. This helps ensure safety and improve the production efficiency of IoT in mining.
2023, 45(10): 3558-3567.
doi: 10.11999/JEIT221017
Abstract:
In order to improve the accuracy of light-weight Convolutional Neural Networks (CNN) in the classification task of Remote Sensing Images (RSI) scene, a Double Knowledge Distillation (DKD) model combined with Dual-Attention (DA) and Spatial Structure (SS) is designed in this paper. First, new DA and SS modules are constructed and introduced into ResNet101 and lightweight CNN designed as teacher and student networks respectively. Then, a DA distillation loss function is constructed to transfer DA knowledge from teacher network to student network, so as to enhance their ability to extract local features from RSI. Finally, constructing a SS distillation loss function, migrating the semantic extraction ability in the teacher network to the student network in the form of a spatial structure to enhance its ability to express the high -level semantics of the RSI. The experimental results based on two standard data sets AID and NWPU-45 show that the performance of the student network after knowledge distillation is improved by 7.57% and 7.28% respectively under the condition of 20% training proportion, and the performance is still better than other methods under the condition of fewer parameters.
In order to improve the accuracy of light-weight Convolutional Neural Networks (CNN) in the classification task of Remote Sensing Images (RSI) scene, a Double Knowledge Distillation (DKD) model combined with Dual-Attention (DA) and Spatial Structure (SS) is designed in this paper. First, new DA and SS modules are constructed and introduced into ResNet101 and lightweight CNN designed as teacher and student networks respectively. Then, a DA distillation loss function is constructed to transfer DA knowledge from teacher network to student network, so as to enhance their ability to extract local features from RSI. Finally, constructing a SS distillation loss function, migrating the semantic extraction ability in the teacher network to the student network in the form of a spatial structure to enhance its ability to express the high -level semantics of the RSI. The experimental results based on two standard data sets AID and NWPU-45 show that the performance of the student network after knowledge distillation is improved by 7.57% and 7.28% respectively under the condition of 20% training proportion, and the performance is still better than other methods under the condition of fewer parameters.
2023, 45(10): 3568-3577.
doi: 10.11999/JEIT221113
Abstract:
In order to improve the high Bit Error Rate(BER) of multi-user Correlated Delay Shift Keying(CDSK) system, a Multi-User Multi-Carrier Correlated Delay Shift Keying system based on Time Slot Transformation (TST-MUMC-CDSK) system is proposed. The scheme uses a permutation matrix to transform the reference signal for the purpose of transmitting multi-user information. At the transmitter, the chaotic signals are copied P times and passed through the time slot converter together with the Hilbert transformed orthogonal signals, resulting in 2N mutually orthogonal chaotic signals to carry 2N user messages, greatly increasing the transmission rate of the system. In this paper, the theoretical BER expressions of the system are derived for the Additive White Gaussian Noise (AWGN) channel and the Rayleigh Fading Channel (RFC), and the correctness of the theoretical derivation is verified by numerical simulation. Simulation results show that the proposed system can save more bit energy with the same BER than other similar systems, and the required signal-to-noise ratio is 1.5 dB lower than that of Noise Reduction Multi-User Correlated Delay Shift Keying(NR_MUCDSK) and 2.6 dB lower than that of Multi-User Correlated Delay Shift Keying with No Intra-Signal Interference(NISI_MU_CDSK) at the same BER of 10-2. The system has good theoretical value and provides a good reference for practical engineering applications.
In order to improve the high Bit Error Rate(BER) of multi-user Correlated Delay Shift Keying(CDSK) system, a Multi-User Multi-Carrier Correlated Delay Shift Keying system based on Time Slot Transformation (TST-MUMC-CDSK) system is proposed. The scheme uses a permutation matrix to transform the reference signal for the purpose of transmitting multi-user information. At the transmitter, the chaotic signals are copied P times and passed through the time slot converter together with the Hilbert transformed orthogonal signals, resulting in 2N mutually orthogonal chaotic signals to carry 2N user messages, greatly increasing the transmission rate of the system. In this paper, the theoretical BER expressions of the system are derived for the Additive White Gaussian Noise (AWGN) channel and the Rayleigh Fading Channel (RFC), and the correctness of the theoretical derivation is verified by numerical simulation. Simulation results show that the proposed system can save more bit energy with the same BER than other similar systems, and the required signal-to-noise ratio is 1.5 dB lower than that of Noise Reduction Multi-User Correlated Delay Shift Keying(NR_MUCDSK) and 2.6 dB lower than that of Multi-User Correlated Delay Shift Keying with No Intra-Signal Interference(NISI_MU_CDSK) at the same BER of 10-2. The system has good theoretical value and provides a good reference for practical engineering applications.
2023, 45(10): 3578-3586.
doi: 10.11999/JEIT220991
Abstract:
Object detection in remote sensing image is one of the hot research topics in the field of remote sensing. In order to adapt to complex backgrounds and multi-directional objects in remote sensing images, the mainstream object detection model uses rotation detection method. However, most of positioning losses used for rotation detection generally has the problem that its trend is inconsistent with the trend of SkewIoU(Skew Intersection-over-Union). To solve this problem, a new bidirectional attenuation loss for rotating object detection is designed. Specifically, this method simulates SkewIoU by Gaussian product, and attenuates the product from two directions according to the deviation of the predicted position. The bidirectional attenuation loss has stronger trend-level alignment with SkewIoU and works better compared with other methods, thanks to its ability to reflect the SkewIoU change caused by position deviation. Experiments on DOTAv1.0 show the effectiveness of this method of various loss forms and different accuracy conditions.
Object detection in remote sensing image is one of the hot research topics in the field of remote sensing. In order to adapt to complex backgrounds and multi-directional objects in remote sensing images, the mainstream object detection model uses rotation detection method. However, most of positioning losses used for rotation detection generally has the problem that its trend is inconsistent with the trend of SkewIoU(Skew Intersection-over-Union). To solve this problem, a new bidirectional attenuation loss for rotating object detection is designed. Specifically, this method simulates SkewIoU by Gaussian product, and attenuates the product from two directions according to the deviation of the predicted position. The bidirectional attenuation loss has stronger trend-level alignment with SkewIoU and works better compared with other methods, thanks to its ability to reflect the SkewIoU change caused by position deviation. Experiments on DOTAv1.0 show the effectiveness of this method of various loss forms and different accuracy conditions.
2023, 45(10): 3587-3593.
doi: 10.11999/JEIT220967
Abstract:
For the Doppler frequency offset of Link-16 data link terminal platform in low Signal-Noise Ratio(SNR) and high-speed moving scene, a new data structure is designed, and a step-by-step frequency offset estimation algorithm combining frequency domain transform, and the Cramer-Rao Low Bound(CRLB) and time-domain autocorrelation is proposed on this basis. First according to the basic idea, the received signal is autocorrelated, and then the maximum index is found through frequency domain transformation. Combined with the correlation factor, the coarse estimation value of frequency offset is obtained. Then, the received signal is finely estimated by using improved L&R algorithm in time domain, and the final frequency offset estimation value is obtained according to the two-step estimation algorithm. The algorithm is simulated by Monte Carlo experiment. The simulation results show that compared with the traditional frequency offset estimation algorithm, the normalized mean square error of the algorithm is closer to CRLB, and when the Doppler frequency offset is [–20 kHz, 20 kHz], the estimation accuracy can reach 10–5. In low SNR environment, the algorithm can achieve ideal estimation effect, which is suitable for Link-16 data link communication.
For the Doppler frequency offset of Link-16 data link terminal platform in low Signal-Noise Ratio(SNR) and high-speed moving scene, a new data structure is designed, and a step-by-step frequency offset estimation algorithm combining frequency domain transform, and the Cramer-Rao Low Bound(CRLB) and time-domain autocorrelation is proposed on this basis. First according to the basic idea, the received signal is autocorrelated, and then the maximum index is found through frequency domain transformation. Combined with the correlation factor, the coarse estimation value of frequency offset is obtained. Then, the received signal is finely estimated by using improved L&R algorithm in time domain, and the final frequency offset estimation value is obtained according to the two-step estimation algorithm. The algorithm is simulated by Monte Carlo experiment. The simulation results show that compared with the traditional frequency offset estimation algorithm, the normalized mean square error of the algorithm is closer to CRLB, and when the Doppler frequency offset is [–20 kHz, 20 kHz], the estimation accuracy can reach 10–5. In low SNR environment, the algorithm can achieve ideal estimation effect, which is suitable for Link-16 data link communication.
2023, 45(10): 3594-3602.
doi: 10.11999/JEIT221392
Abstract:
In order to reduce the candidate flip bit set size when using Fast Simplified Successive Cancellation Flip(Fast-SSC-Flip) decoding algorithm and decrease the search complexity, a kind of Fast-SSC-Flip decoding algorithm based on critical flip set for polar code is proposed. Based on the fact that the first decoding error information bit in the Fast Simplified Successive Cancellation (Fast-SSC) decoding process is highly likely to fall into the Critical Set (CS), and the candidate bits in the Fast-SSC-Flip decoding algorithm are all codeword bits, the proposed algorithm uses the polar code generator matrix to obtain the corresponding codeword bits of the information bits in the CS, and constructs a Critical Flip Set (CFS) as the candidate flip bit set. Experimental results show that, under the same candidate bit reliability measurement criteria, when the code length N = 1 024 and the code rate R = 0.5, the proposed Fast-SSC-Flip decoding algorithm based on critical flip set reduces significantly the candidate flip bit set size without sacrificing decoding performance compared to the traditional Fast-SSC-Flip algorithm; Compared to the New Fast Simplified Successive Cancellation Flip( N-Fast-SSC-Flip) algorithm, the candidate flip bit set decreased by at least 77.93% while maintaining similar decoding performance.
In order to reduce the candidate flip bit set size when using Fast Simplified Successive Cancellation Flip(Fast-SSC-Flip) decoding algorithm and decrease the search complexity, a kind of Fast-SSC-Flip decoding algorithm based on critical flip set for polar code is proposed. Based on the fact that the first decoding error information bit in the Fast Simplified Successive Cancellation (Fast-SSC) decoding process is highly likely to fall into the Critical Set (CS), and the candidate bits in the Fast-SSC-Flip decoding algorithm are all codeword bits, the proposed algorithm uses the polar code generator matrix to obtain the corresponding codeword bits of the information bits in the CS, and constructs a Critical Flip Set (CFS) as the candidate flip bit set. Experimental results show that, under the same candidate bit reliability measurement criteria, when the code length N = 1 024 and the code rate R = 0.5, the proposed Fast-SSC-Flip decoding algorithm based on critical flip set reduces significantly the candidate flip bit set size without sacrificing decoding performance compared to the traditional Fast-SSC-Flip algorithm; Compared to the New Fast Simplified Successive Cancellation Flip( N-Fast-SSC-Flip) algorithm, the candidate flip bit set decreased by at least 77.93% while maintaining similar decoding performance.
2023, 45(10): 3603-3611.
doi: 10.11999/JEIT220946
Abstract:
Reconfigurable Intelligent Surface (RIS) can be regarded as a ‘relay’ with special functions in communication network. It can cooperate with Non-Orthogonal Multiple Access (NOMA) system to construct a coordinated information transmission scheme. Considering the different Quality of Service (QoS) requirements of different user devices in the future Internet of Things (IoT) scenarios, a RIS-assisted multi-user NOMA communication system model is proposed. According to the different QoS requirements of two types of users (information users and energy users), a power allocation method based on iterative optimization is designed. This method minimizes the total transmit power of the system by jointly designing the phase-shift matrix of RIS, the beamforming of base station and the order of successive interference cancellation in NOMA system, so as to reduce comprehensively the energy consumption of base station. Simulation results show that compared with the scenario without RIS, RIS-assisted NOMA system can effectively reduce the energy consumption of base station. In the case with RIS, the energy consumption of the proposed power allocation method is significantly lower than that of random phase selection at RIS and zero-forcing beamforming at base station.
Reconfigurable Intelligent Surface (RIS) can be regarded as a ‘relay’ with special functions in communication network. It can cooperate with Non-Orthogonal Multiple Access (NOMA) system to construct a coordinated information transmission scheme. Considering the different Quality of Service (QoS) requirements of different user devices in the future Internet of Things (IoT) scenarios, a RIS-assisted multi-user NOMA communication system model is proposed. According to the different QoS requirements of two types of users (information users and energy users), a power allocation method based on iterative optimization is designed. This method minimizes the total transmit power of the system by jointly designing the phase-shift matrix of RIS, the beamforming of base station and the order of successive interference cancellation in NOMA system, so as to reduce comprehensively the energy consumption of base station. Simulation results show that compared with the scenario without RIS, RIS-assisted NOMA system can effectively reduce the energy consumption of base station. In the case with RIS, the energy consumption of the proposed power allocation method is significantly lower than that of random phase selection at RIS and zero-forcing beamforming at base station.
2023, 45(10): 3612-3621.
doi: 10.11999/JEIT221232
Abstract:
As the channel status information is difficult to obtain in Reconfigurable Intelligent Surface(RIS) aided millimeter wave communication. The radio frequency chain is provided in some passive elements of RIS to assist channel estimation and estimate the channel between Base Station(BS)/User Equipment(UE) and RIS. Inspired by the structure, a low complexity channel estimation method is proposed. First, the two-dimensional angle estimation problem is transformed into two one-dimensional angle estimation Semi-positive definite programming (SDP) problem by the Atomic Norm Minimization(ANM) method.Second, an Alternating Direction Method of Multipliers(ADMM) algorithm is proposed to solve the Semi-Definite Programming (SDP) problem. Based on the ADMM algorithm, the algorithm applies momentum gradient descent method to avoid the inverse operation of the matrix, and improves the channel estimation accuracy by joint optimizing between a iteration step and channel parameters. Finally, the path gain estimation is obtained by using the signal angle and channel matrix. Simulation results show that the proposed algorithm has better channel estimation performance, and based on the system parameter set, the proposed algorithm has a low complexity.
As the channel status information is difficult to obtain in Reconfigurable Intelligent Surface(RIS) aided millimeter wave communication. The radio frequency chain is provided in some passive elements of RIS to assist channel estimation and estimate the channel between Base Station(BS)/User Equipment(UE) and RIS. Inspired by the structure, a low complexity channel estimation method is proposed. First, the two-dimensional angle estimation problem is transformed into two one-dimensional angle estimation Semi-positive definite programming (SDP) problem by the Atomic Norm Minimization(ANM) method.Second, an Alternating Direction Method of Multipliers(ADMM) algorithm is proposed to solve the Semi-Definite Programming (SDP) problem. Based on the ADMM algorithm, the algorithm applies momentum gradient descent method to avoid the inverse operation of the matrix, and improves the channel estimation accuracy by joint optimizing between a iteration step and channel parameters. Finally, the path gain estimation is obtained by using the signal angle and channel matrix. Simulation results show that the proposed algorithm has better channel estimation performance, and based on the system parameter set, the proposed algorithm has a low complexity.
2023, 45(10): 3622-3630.
doi: 10.11999/JEIT221413
Abstract:
In recent years, quaternion theory has become a research hotspot for scholars and has been applied to many fields. In this paper, the quadrature polarization channel equalization problem is studied based on quaternion adaptive filtering algorithm. In order to solve the phase ambiguity problem of Quaternion Constant Modulus Algorithm (QCMA), a concurrent quaternion Direct Decision constant modulus algorithm (QCMA+DD-QLMS) is proposed by combining QCMA algorithm with Quaternion Least Mean Square (QLMS) algorithm. Based on the gradient operation rules of Generalized Hamilton-Real (GHR), the new algorithm is theoretically deduced and simulated by MATLAB. The simulation results show that the algorithm proposed in this paper can not only solve the phase ambiguity problem of the QCMA, but also has smaller steady-state Mean Square Error (MSE).
In recent years, quaternion theory has become a research hotspot for scholars and has been applied to many fields. In this paper, the quadrature polarization channel equalization problem is studied based on quaternion adaptive filtering algorithm. In order to solve the phase ambiguity problem of Quaternion Constant Modulus Algorithm (QCMA), a concurrent quaternion Direct Decision constant modulus algorithm (QCMA+DD-QLMS) is proposed by combining QCMA algorithm with Quaternion Least Mean Square (QLMS) algorithm. Based on the gradient operation rules of Generalized Hamilton-Real (GHR), the new algorithm is theoretically deduced and simulated by MATLAB. The simulation results show that the algorithm proposed in this paper can not only solve the phase ambiguity problem of the QCMA, but also has smaller steady-state Mean Square Error (MSE).
2023, 45(10): 3631-3639.
doi: 10.11999/JEIT221102
Abstract:
The uplink Physical Layer Security (PLS) of large-scale Non-Orthogonal Multiple Access (NOMA)-based Fog access networks with a Reconfigurable Intelligent Surface (RIS)-assisted Cell-Free (CF) framework is investigated in this paper. The spatial effects of Fog-Access Point (F-AP) locations are introduced into the RIS reflecting model design by invoking stochastic geometry and Power Domain (PD) multiplexing schemes, this spatial-repulsion-based co-design approach is proposed to enhance the PLS of the considered system. This system is modeled with reference to a receiver-transmitter pair, the Fisher-Snedecor\begin{document}$\mathcal{F}$\end{document} \begin{document}$10.5\% $\end{document}
The uplink Physical Layer Security (PLS) of large-scale Non-Orthogonal Multiple Access (NOMA)-based Fog access networks with a Reconfigurable Intelligent Surface (RIS)-assisted Cell-Free (CF) framework is investigated in this paper. The spatial effects of Fog-Access Point (F-AP) locations are introduced into the RIS reflecting model design by invoking stochastic geometry and Power Domain (PD) multiplexing schemes, this spatial-repulsion-based co-design approach is proposed to enhance the PLS of the considered system. This system is modeled with reference to a receiver-transmitter pair, the Fisher-Snedecor
A Low Complexity Algorithm for Intelligent Reflective Surface-assisted Tera Hertz Channel Estimation
2023, 45(10): 3640-3647.
doi: 10.11999/JEIT221476
Abstract:
Channel estimation is a major challenge in Tera Hertz (THz) communication assisted by Intelligent Reflecting Surfaces (IRS). To reduce the problem of large channel estimation pilot overhead caused by an increase in the number of receiving/transmitting antennas and IRS reflecting elements, a channel estimation algorithm based on Canonical decomposition/Parallel factors (CP) decomposition is proposed. Firstly, based on the analysis of the channel characteristics, the IRS elements are designed by grouping. Then, the wireless communication channel assisted by IRS is expressed in a unified mathematical expression. Subsequently, the received signal matrix is constructed into a three-dimensional tensor by utilizing the inherent low-rank structure of the THz channel with multiple antennas. The tensor is decomposed using the CP decomposition algorithm, and the channel parameters are estimated using correlation. Monte Carlo simulations show that the proposed algorithm has a performance improvement of around 4.28 dB and 7.12 dB compared to the benchmark algorithm under the same channel transmission conditions, and has a lower computational complexity.
Channel estimation is a major challenge in Tera Hertz (THz) communication assisted by Intelligent Reflecting Surfaces (IRS). To reduce the problem of large channel estimation pilot overhead caused by an increase in the number of receiving/transmitting antennas and IRS reflecting elements, a channel estimation algorithm based on Canonical decomposition/Parallel factors (CP) decomposition is proposed. Firstly, based on the analysis of the channel characteristics, the IRS elements are designed by grouping. Then, the wireless communication channel assisted by IRS is expressed in a unified mathematical expression. Subsequently, the received signal matrix is constructed into a three-dimensional tensor by utilizing the inherent low-rank structure of the THz channel with multiple antennas. The tensor is decomposed using the CP decomposition algorithm, and the channel parameters are estimated using correlation. Monte Carlo simulations show that the proposed algorithm has a performance improvement of around 4.28 dB and 7.12 dB compared to the benchmark algorithm under the same channel transmission conditions, and has a lower computational complexity.
2023, 45(10): 3648-3656.
doi: 10.11999/JEIT221398
Abstract:
In recent years, Unmanned Aerial Vehicles (UAVs) have been widely used in post-disaster rescue by virtue of their mobility and flexibility. Considering the scenario that a survey UAV performs tasks using the emergency communication network, in order to extend the overall endurance of the emergency communication network, in this paper, the energy efficiency of the system is maximize by jointly optimizing the relay selection and flight trajectory of the UAV. In addition, the available communication energy of the relay UAVs as well as the maximum flight speed and real-time communication quality of the survey UAV are also considered in the optimization. The resultant Nondeterministic Polynomial hard (NP-hard) optimization problem is approximately solved using an alternate algorithm, which consists of successive convex approximation and tabu search algorithm. The alternate algorithm splits the original problem into two subproblems and solves them alternately to obtain the approximate optimal solution of the optimization problem. Simulation results show that the proposed algorithm has a desirable convergence and can significantly improve the energy efficiency of the system. The performance of the proposed algorithm is improved by 31.1% and 28.2% compared to the benchmark schemes of relay or trajectory optimization.
In recent years, Unmanned Aerial Vehicles (UAVs) have been widely used in post-disaster rescue by virtue of their mobility and flexibility. Considering the scenario that a survey UAV performs tasks using the emergency communication network, in order to extend the overall endurance of the emergency communication network, in this paper, the energy efficiency of the system is maximize by jointly optimizing the relay selection and flight trajectory of the UAV. In addition, the available communication energy of the relay UAVs as well as the maximum flight speed and real-time communication quality of the survey UAV are also considered in the optimization. The resultant Nondeterministic Polynomial hard (NP-hard) optimization problem is approximately solved using an alternate algorithm, which consists of successive convex approximation and tabu search algorithm. The alternate algorithm splits the original problem into two subproblems and solves them alternately to obtain the approximate optimal solution of the optimization problem. Simulation results show that the proposed algorithm has a desirable convergence and can significantly improve the energy efficiency of the system. The performance of the proposed algorithm is improved by 31.1% and 28.2% compared to the benchmark schemes of relay or trajectory optimization.
2023, 45(10): 3657-3666.
doi: 10.11999/JEIT221160
Abstract:
The WiFi fingerprint based on Channel State Information (CSI) data can be used for indoor positioning. Compared to Received Signal Strength Indicator (RSSI) data, CSI has a higher granularity of data information and can be obtained over multiple subcarriers. Better results can be achieved when using CSI data for indoor localization. However, regardless of whether RSSI or CSI signals are used, the indoor environment often changes after a period of time during the deployment of indoor localization, and the fingerprint database based on the test data often deteriorates or even becomes invalid. In this paper, using a transfer learning algorithm to establish a fingerprint database for indoor positioning is proposed. The advantage of transfer learning is that it can use less data to obtain better transfer training results. Transfer learning is used to migrate the prediction of fingerprint database, the life cycle of fingerprint database is prolonged, and robustness in indoor positioning is improved. The indoor positioning accuracy is maintained at 98% after one week and 97% after two weeks. At the same cost, the life cycle and positioning accuracy of the proposed model are higher than Long Short-Term Memory (LSTM), Convolutional Neural Networks (CNN), Support Vector Machine (SVM), Deep Neural Networks (DNN), and other positioning systems.
The WiFi fingerprint based on Channel State Information (CSI) data can be used for indoor positioning. Compared to Received Signal Strength Indicator (RSSI) data, CSI has a higher granularity of data information and can be obtained over multiple subcarriers. Better results can be achieved when using CSI data for indoor localization. However, regardless of whether RSSI or CSI signals are used, the indoor environment often changes after a period of time during the deployment of indoor localization, and the fingerprint database based on the test data often deteriorates or even becomes invalid. In this paper, using a transfer learning algorithm to establish a fingerprint database for indoor positioning is proposed. The advantage of transfer learning is that it can use less data to obtain better transfer training results. Transfer learning is used to migrate the prediction of fingerprint database, the life cycle of fingerprint database is prolonged, and robustness in indoor positioning is improved. The indoor positioning accuracy is maintained at 98% after one week and 97% after two weeks. At the same cost, the life cycle and positioning accuracy of the proposed model are higher than Long Short-Term Memory (LSTM), Convolutional Neural Networks (CNN), Support Vector Machine (SVM), Deep Neural Networks (DNN), and other positioning systems.
2023, 45(10): 3667-3675.
doi: 10.11999/JEIT221412
Abstract:
Network virtualization allows multiple virtual networks to coexist on the same physical infrastructure, facilitating the incremental deployment of future network technology. However, the current programmable data plane provides exclusive data plane abstract that is difficult to support multiple network functions simultaneously. A Virtualized P4-based Programmable Data Plane architecture with Parallel Pipeline (VirtP6) is proposed, which allows multiple isolated virtual network functions to run on a single physical device. The single pipeline structure of the programmable data plane is changed. Then, multiple parallel packet processing pipelines are ensured to realize the virtualization of the programmable data plane, and resource isolation, traffic isolation and access isolation between different virtual network functions are introduced. Finally, the virtualization overhead, isolation, scalability, and network applicability of VirtP6 are evaluated. Experimental results show that compared with HyperP4, VirtP6 reduces greatly virtualization overhead, reduces latency by 68%, improves throughput by 75%, and has good isolation and scalability.
Network virtualization allows multiple virtual networks to coexist on the same physical infrastructure, facilitating the incremental deployment of future network technology. However, the current programmable data plane provides exclusive data plane abstract that is difficult to support multiple network functions simultaneously. A Virtualized P4-based Programmable Data Plane architecture with Parallel Pipeline (VirtP6) is proposed, which allows multiple isolated virtual network functions to run on a single physical device. The single pipeline structure of the programmable data plane is changed. Then, multiple parallel packet processing pipelines are ensured to realize the virtualization of the programmable data plane, and resource isolation, traffic isolation and access isolation between different virtual network functions are introduced. Finally, the virtualization overhead, isolation, scalability, and network applicability of VirtP6 are evaluated. Experimental results show that compared with HyperP4, VirtP6 reduces greatly virtualization overhead, reduces latency by 68%, improves throughput by 75%, and has good isolation and scalability.
2023, 45(10): 3676-3684.
doi: 10.11999/JEIT221137
Abstract:
To reflect better the inherent characteristics of chaotic systems, an adaptive noise reduction algorithm for chaotic signals based on wavelet packet transform is proposed. Firstly, the best decomposition level is determined according to the different correlation of wavelet packet coefficients in different decomposition scales, while the optimal wavelet packet basis is obtained with the logarithmic energy entropy as the cost function. Then, the approximate coefficients are projected in the local neighborhood and the detail coefficients are adaptively selected with the gradient descent algorithm in neural network. By minimizing the loss function, the influence of noises on chaotic signals is reduced to the greatest extent. Finally, simulations on the state variables originating from Rossler chaotic model verify the denoising superiority of the proposed algorithm for the chaotic signals.
To reflect better the inherent characteristics of chaotic systems, an adaptive noise reduction algorithm for chaotic signals based on wavelet packet transform is proposed. Firstly, the best decomposition level is determined according to the different correlation of wavelet packet coefficients in different decomposition scales, while the optimal wavelet packet basis is obtained with the logarithmic energy entropy as the cost function. Then, the approximate coefficients are projected in the local neighborhood and the detail coefficients are adaptively selected with the gradient descent algorithm in neural network. By minimizing the loss function, the influence of noises on chaotic signals is reduced to the greatest extent. Finally, simulations on the state variables originating from Rossler chaotic model verify the denoising superiority of the proposed algorithm for the chaotic signals.
2023, 45(10): 3685-3693.
doi: 10.11999/JEIT221165
Abstract:
To solve the network topology management difficulties caused by the rapid movement of Unmanned Aerial Vehicles (UAVs) in the Flying Ad hoc NETworks (FANET), an adaptive hummingbird algorithm is proposed to optimize the communication topology, which considers differences in available channels caused by the change of UAVs position in practical applications. Firstly, a UAV topology model for the clustered structure is established, and an optimization problem is formed to minimize the number of clusters, load deviation, and cluster mobility. Secondly, by adjusting the foraging action of artificial hummingbirds and adding disturbance variation, an ADaptive Hummingbird Algorithm (ADHA) with a stronger search ability is proposed. Thirdly, a reasonable hummingbird individual coding method is designed, and the decision-making process of topology optimization is transformed into the optimization process of ADHA. Finally, the convergence of the proposed algorithm is verified by simulation, and it is compared with other topology optimization methods based on other swarm intelligence optimization algorithms. The experimental results show that the topology optimization strategy obtained by the proposed algorithm can not only effectively reduce the number of clusters in the network topology, but also obtain clusters with balanced load and stable structure.
To solve the network topology management difficulties caused by the rapid movement of Unmanned Aerial Vehicles (UAVs) in the Flying Ad hoc NETworks (FANET), an adaptive hummingbird algorithm is proposed to optimize the communication topology, which considers differences in available channels caused by the change of UAVs position in practical applications. Firstly, a UAV topology model for the clustered structure is established, and an optimization problem is formed to minimize the number of clusters, load deviation, and cluster mobility. Secondly, by adjusting the foraging action of artificial hummingbirds and adding disturbance variation, an ADaptive Hummingbird Algorithm (ADHA) with a stronger search ability is proposed. Thirdly, a reasonable hummingbird individual coding method is designed, and the decision-making process of topology optimization is transformed into the optimization process of ADHA. Finally, the convergence of the proposed algorithm is verified by simulation, and it is compared with other topology optimization methods based on other swarm intelligence optimization algorithms. The experimental results show that the topology optimization strategy obtained by the proposed algorithm can not only effectively reduce the number of clusters in the network topology, but also obtain clusters with balanced load and stable structure.
2023, 45(10): 3694-3701.
doi: 10.11999/JEIT221406
Abstract:
Considering the problems of high spectrum resource consumption and high traffic blocking probability during the time-varying traffic’s survivability transmission in Elastic Optical Networks(EONs), a Time-varying Traffic Shared Protection method based on Spectrum Window Sliding (TTSP-SWS) is proposed in this paper. In TTSP-SWS, the protection lightpath with the minimum cost including the load weight of available spectrum block and the sharing degree of protected spectrum block is selected as the protection lightpath for time-varying traffic. By sliding the spectrum window on the protection path, the spectrum window with high sharing degree is allocated to time-varying traffic. When the required bandwidth of time-varying traffic changes, a spectrum window sliding-based survivable spectrum expansion or compressed strategy is adopted to adjust the allocated bandwidth. Simulation results show that TTSP-SWS can reduce traffic blocking ratio and protection resources’ redundancy.
Considering the problems of high spectrum resource consumption and high traffic blocking probability during the time-varying traffic’s survivability transmission in Elastic Optical Networks(EONs), a Time-varying Traffic Shared Protection method based on Spectrum Window Sliding (TTSP-SWS) is proposed in this paper. In TTSP-SWS, the protection lightpath with the minimum cost including the load weight of available spectrum block and the sharing degree of protected spectrum block is selected as the protection lightpath for time-varying traffic. By sliding the spectrum window on the protection path, the spectrum window with high sharing degree is allocated to time-varying traffic. When the required bandwidth of time-varying traffic changes, a spectrum window sliding-based survivable spectrum expansion or compressed strategy is adopted to adjust the allocated bandwidth. Simulation results show that TTSP-SWS can reduce traffic blocking ratio and protection resources’ redundancy.
2023, 45(10): 3702-3709.
doi: 10.11999/JEIT221223
Abstract:
To overcome the limitation of the Federated Learning (FL) when the data and model of each client are all heterogenous and improve the accuracy, a personalized Federated learning algorithm with Collation game and Knowledge distillation (pFedCK) is proposed. Firstly, each client uploads its soft-predict on public dataset and downloads the most correlative of the k soft-predict. Then, the Shapley Value (SV) from collation game is applied to measure the multi-wise influences among clients and their marginal contribution to others on personalized learning performance is quantified. Lastly, each client identify it’s optimal coalition and then the Knowledge Distillation (KD) is used to local model and local training is conduct on the privacy dataset. The results show that compared with the state-of-the-art algorithm, this approach can achieve superior personalized accuracy and can improve by about 10%.
To overcome the limitation of the Federated Learning (FL) when the data and model of each client are all heterogenous and improve the accuracy, a personalized Federated learning algorithm with Collation game and Knowledge distillation (pFedCK) is proposed. Firstly, each client uploads its soft-predict on public dataset and downloads the most correlative of the k soft-predict. Then, the Shapley Value (SV) from collation game is applied to measure the multi-wise influences among clients and their marginal contribution to others on personalized learning performance is quantified. Lastly, each client identify it’s optimal coalition and then the Knowledge Distillation (KD) is used to local model and local training is conduct on the privacy dataset. The results show that compared with the state-of-the-art algorithm, this approach can achieve superior personalized accuracy and can improve by about 10%.
2023, 45(10): 3710-3718.
doi: 10.11999/JEIT220131
Abstract:
Due to the limited actual network bandwidth, the communication between clients and the central server is a main bottleneck of federated learning. To reduce the communication cost, a communication-efficient Federated learning algorithm is proposed by introducing the Event Triggered mechanism (FedET). Firstly, the clients determine whether to send the current model to the central server through using the event-triggered mechanism. Then, the central server aggregates models based on the information received. In particular, at each communication round, after finishing the local model training, the clients compare the model update with the trigger threshold, and if the communication is triggered, the transmitted information is compressed and sent to the central server. Furthermore, for smooth objective functions which satisfy convex, PL (Polyak-Łojasiewicz) condition and non-convex, respectively, this paper analyzes the convergence of the proposed algorithm and presents the proof. Finally, simulation experiments are implemented on two standard datasets. Simulation results verify the feasibility and effectiveness of the proposed algorithm.
Due to the limited actual network bandwidth, the communication between clients and the central server is a main bottleneck of federated learning. To reduce the communication cost, a communication-efficient Federated learning algorithm is proposed by introducing the Event Triggered mechanism (FedET). Firstly, the clients determine whether to send the current model to the central server through using the event-triggered mechanism. Then, the central server aggregates models based on the information received. In particular, at each communication round, after finishing the local model training, the clients compare the model update with the trigger threshold, and if the communication is triggered, the transmitted information is compressed and sent to the central server. Furthermore, for smooth objective functions which satisfy convex, PL (Polyak-Łojasiewicz) condition and non-convex, respectively, this paper analyzes the convergence of the proposed algorithm and presents the proof. Finally, simulation experiments are implemented on two standard datasets. Simulation results verify the feasibility and effectiveness of the proposed algorithm.
2023, 45(10): 3719-3728.
doi: 10.11999/JEIT220959
Abstract:
Considering the problem that the class imbalance of traffic dataset limits the performance of the model to the minority class attack traffic, a traffic anomaly detection method based on the joint model of attention mechanism and One-Dimensional Convolutional Neural Network - Bidirectional Long Short Term Memory (1DCNN-BiLSTM) is proposed. First, in the data preprocessing, the BorderlineSMOTE method is used to preprocess the imbalanced traffic training data, so that the quantities of different categories are balanced, which is helpful for the model to train various types fully. Then, the joint model of attention mechanism and 1DCNN-BiLSTM is designed to extract the local and long-distance sequence features of the traffic data. The features useful for classification are assigned weights according to their importance through the attention mechanism, which makes the model improve the detection rate of attack classes. Experimental results show that the proposed method has the highest accuracy for NSL-KDD and CICIDS2017 datasets (up to 93.17% and 98.65%). The proposed method improves the detection rate of User to Root(U2R) attack traffic in NSL-KDD dataset by at least 13.70%, which proves the effectiveness of the proposed method in improving the detection rate of minority attack traffic.
Considering the problem that the class imbalance of traffic dataset limits the performance of the model to the minority class attack traffic, a traffic anomaly detection method based on the joint model of attention mechanism and One-Dimensional Convolutional Neural Network - Bidirectional Long Short Term Memory (1DCNN-BiLSTM) is proposed. First, in the data preprocessing, the BorderlineSMOTE method is used to preprocess the imbalanced traffic training data, so that the quantities of different categories are balanced, which is helpful for the model to train various types fully. Then, the joint model of attention mechanism and 1DCNN-BiLSTM is designed to extract the local and long-distance sequence features of the traffic data. The features useful for classification are assigned weights according to their importance through the attention mechanism, which makes the model improve the detection rate of attack classes. Experimental results show that the proposed method has the highest accuracy for NSL-KDD and CICIDS2017 datasets (up to 93.17% and 98.65%). The proposed method improves the detection rate of User to Root(U2R) attack traffic in NSL-KDD dataset by at least 13.70%, which proves the effectiveness of the proposed method in improving the detection rate of minority attack traffic.
2023, 45(10): 3729-3736.
doi: 10.11999/JEIT221038
Abstract:
Gimli and Xoodoo are large state lightweight block ciphers that have many advantages such as fewer logic gates, low power consumption and fast encryption, and have attracted much attention on the industry. Both are based on 384 bit permutation, while the large state can lead to the increase of difficulty of security analysis. In this paper, the equivalent representations of the AND, OR and S-boxes operations are introduced. And the automatic search model of the impossible differential distinguisher of Gimli and Xoodoo are constructed. Furthermore, a new technique based on "bisection method" is proposed to detect the contradiction for the impossible differential distinguisher, which is used to verify the correctness of the distinguisher. The results show that the impossible differential distinguishers of 10-round Gimli and 4-round Xoodoo are obtained and verified in this paper. Especially, the new impossible differential distinguisher of Gimli is increased by 3 rounds compared with the existing results.
Gimli and Xoodoo are large state lightweight block ciphers that have many advantages such as fewer logic gates, low power consumption and fast encryption, and have attracted much attention on the industry. Both are based on 384 bit permutation, while the large state can lead to the increase of difficulty of security analysis. In this paper, the equivalent representations of the AND, OR and S-boxes operations are introduced. And the automatic search model of the impossible differential distinguisher of Gimli and Xoodoo are constructed. Furthermore, a new technique based on "bisection method" is proposed to detect the contradiction for the impossible differential distinguisher, which is used to verify the correctness of the distinguisher. The results show that the impossible differential distinguishers of 10-round Gimli and 4-round Xoodoo are obtained and verified in this paper. Especially, the new impossible differential distinguisher of Gimli is increased by 3 rounds compared with the existing results.
2023, 45(10): 3737-3744.
doi: 10.11999/JEIT221282
Abstract:
The LEA algorithm is a software-oriented lightweight encryption algorithm, which became the ISO/IEC international standard lightweight encryption algorithm in 2019. It has the advantages of fast encryption and less computing resources. The differential probability is calculated based on multiple paths with the same input-output difference, 13 and 14 rounds of key recovery attacks of LEA-128 are given for the first time. Using the early abort technology, one round is added after the 12-round and 13-round differential characteristic, and a total of 96 bit keys are recovered. The 13-round key recovery attack has a data complexity of 298 plaintext and a time complexity of 286.7 times of 13 rounds of LEA-128 decryption; the 14-round key recovery attack has a data complexity of 2118 plaintext and a time complexity of 2110.6 times of 14 rounds of LEA-128 decryption.
The LEA algorithm is a software-oriented lightweight encryption algorithm, which became the ISO/IEC international standard lightweight encryption algorithm in 2019. It has the advantages of fast encryption and less computing resources. The differential probability is calculated based on multiple paths with the same input-output difference, 13 and 14 rounds of key recovery attacks of LEA-128 are given for the first time. Using the early abort technology, one round is added after the 12-round and 13-round differential characteristic, and a total of 96 bit keys are recovered. The 13-round key recovery attack has a data complexity of 298 plaintext and a time complexity of 286.7 times of 13 rounds of LEA-128 decryption; the 14-round key recovery attack has a data complexity of 2118 plaintext and a time complexity of 2110.6 times of 14 rounds of LEA-128 decryption.
2023, 45(10): 3745-3751.
doi: 10.11999/JEIT221003
Abstract:
Currently, the security analysis of lightweight block ciphers by using deep learning is becoming a new research hotspot. At the Crypto2019, Gohr first applied deep learning to the security analysis of block ciphers, the high-accuracy neural distinguisher is constructed, which used convolutional neural networks to learn the ciphertext distribution of the given input differentials. LBlock is a lightweight block cipher with excellent software and hardware implementation efficiency, which attracted extensive attention from scholars since its publication. In this paper, with the application of the residual network, a round-reduced neural differential distinguisher of LBlock is constructed, in which the accuracy of the 7-round and 8-round distinguishers reach 0.999 and 0.946, respectively. Moreover, based on the 9-round neural distinguisher, a key recovery attack scheme against 11-round LBlock is proposed. Experiment results show that under the case of the number of iteration rounds of the algorithm is small, the scheme need not consider the S-box separately. Compared with the traditional attack schemes, the new scheme is not only simpler and easy to be implemented, but also possess great advantages on data complexity and time complexity.
Currently, the security analysis of lightweight block ciphers by using deep learning is becoming a new research hotspot. At the Crypto2019, Gohr first applied deep learning to the security analysis of block ciphers, the high-accuracy neural distinguisher is constructed, which used convolutional neural networks to learn the ciphertext distribution of the given input differentials. LBlock is a lightweight block cipher with excellent software and hardware implementation efficiency, which attracted extensive attention from scholars since its publication. In this paper, with the application of the residual network, a round-reduced neural differential distinguisher of LBlock is constructed, in which the accuracy of the 7-round and 8-round distinguishers reach 0.999 and 0.946, respectively. Moreover, based on the 9-round neural distinguisher, a key recovery attack scheme against 11-round LBlock is proposed. Experiment results show that under the case of the number of iteration rounds of the algorithm is small, the scheme need not consider the S-box separately. Compared with the traditional attack schemes, the new scheme is not only simpler and easy to be implemented, but also possess great advantages on data complexity and time complexity.
2023, 45(10): 3752-3758.
doi: 10.11999/JEIT221212
Abstract:
Network coding allows the intermediate nodes to encode and forward the data packets, so it improves the system robustness and saves the network resources. Pollution attacks are inevitable for network coding in applications. In view of this, CertificateLess Pairing-Free SignCryption for Network Coding (NC-CLPFSC) is proposed. NC-CLPFSC uses the homomorphic Hash function to resist the pollution attacks; Moreover, it avoids the key escrow and certificate management, and it reduces the computation and communication cost.
Network coding allows the intermediate nodes to encode and forward the data packets, so it improves the system robustness and saves the network resources. Pollution attacks are inevitable for network coding in applications. In view of this, CertificateLess Pairing-Free SignCryption for Network Coding (NC-CLPFSC) is proposed. NC-CLPFSC uses the homomorphic Hash function to resist the pollution attacks; Moreover, it avoids the key escrow and certificate management, and it reduces the computation and communication cost.
2023, 45(10): 3759-3766.
doi: 10.11999/JEIT210717
Abstract:
Focusing on the problems of over-reliance on Third-Party Auditors (TPA), complex key and certificate management and data leakage in traditional cloud data integrity auditing scheme, a decentralized data integrity verification audit scheme based on blockchain and edge computing is proposed. In order to achieve the decentralization of the audit scheme, fog nodes and smart contracts are used to replace the third-party auditors. Using blockchain to design smart contracts to ensure fair transactions among entities. The proofs generated by audit process are stored in the blockchain, which can prevent the dishonest behavior of each entity. The certificateless cryptosystem is introduced to solve the complex key escrow and certificate management problems in the traditional audit scheme. In addition, the access authorization and identity authentication of cloud data users is realized through encrypted accumulators. The analysis results show that this scheme satisfies the robustness of audit and the unforgeability of signatures, and has higher computing performance compared with similar schemes.
Focusing on the problems of over-reliance on Third-Party Auditors (TPA), complex key and certificate management and data leakage in traditional cloud data integrity auditing scheme, a decentralized data integrity verification audit scheme based on blockchain and edge computing is proposed. In order to achieve the decentralization of the audit scheme, fog nodes and smart contracts are used to replace the third-party auditors. Using blockchain to design smart contracts to ensure fair transactions among entities. The proofs generated by audit process are stored in the blockchain, which can prevent the dishonest behavior of each entity. The certificateless cryptosystem is introduced to solve the complex key escrow and certificate management problems in the traditional audit scheme. In addition, the access authorization and identity authentication of cloud data users is realized through encrypted accumulators. The analysis results show that this scheme satisfies the robustness of audit and the unforgeability of signatures, and has higher computing performance compared with similar schemes.
2023, 45(10): 3767-3775.
doi: 10.11999/JEIT221341
Abstract:
The restoration of geometric structural information in Synthetic Aperture Radar (SAR) remote sensing images is of great significance for target recognition and interpretation. Existed Digital Ortho Map (DOM) production methods for SAR images follow the traditional ortho-rectification methods of optical images. The application of directly object image sampling can not eliminate the geometric distortion caused by the layovers of SAR remote sensing images. Therefore, a method for the production of SAR True Digital Ortho Map (TDOM) is proposed in this paper. With the help of high-precision Digital Surface Model (DSM) data, the layover area is extracted through the back-projection from image-space to object-space. The highest intersection of the DSM and the back-projected lines are considered as the object mapping of the coordinates in layover area to generate the single-view SAR TDOM. Finally, single-view SAR TDOMs are fused together for information compensation, and the multi-view SAR TDOM is produced with multiple information. Images obtained from the GF-3 SAR satellite are experimented, and the results indicate that the structural information of the target in the TDOMs produced by the proposed method is more clearly than traditional produced DOMs. The application of the proposed method can effectively improve the efficiency of target recognition and interpretation in SAR images.
The restoration of geometric structural information in Synthetic Aperture Radar (SAR) remote sensing images is of great significance for target recognition and interpretation. Existed Digital Ortho Map (DOM) production methods for SAR images follow the traditional ortho-rectification methods of optical images. The application of directly object image sampling can not eliminate the geometric distortion caused by the layovers of SAR remote sensing images. Therefore, a method for the production of SAR True Digital Ortho Map (TDOM) is proposed in this paper. With the help of high-precision Digital Surface Model (DSM) data, the layover area is extracted through the back-projection from image-space to object-space. The highest intersection of the DSM and the back-projected lines are considered as the object mapping of the coordinates in layover area to generate the single-view SAR TDOM. Finally, single-view SAR TDOMs are fused together for information compensation, and the multi-view SAR TDOM is produced with multiple information. Images obtained from the GF-3 SAR satellite are experimented, and the results indicate that the structural information of the target in the TDOMs produced by the proposed method is more clearly than traditional produced DOMs. The application of the proposed method can effectively improve the efficiency of target recognition and interpretation in SAR images.
2023, 45(10): 3776-3785.
doi: 10.11999/JEIT221072
Abstract:
Ground Penetrating Radar (GPR) is an underground nondestructive detection technology based on electromagnetic wave, which is widely used in municipal engineering, transportation, military and other fields. In the process of data acquisition, due to the coupling between transmitting antenna and receiving antenna, scattering from undulating ground and the complexity of underground random media, there is usually clutter in the GPR B-scan, which affects seriously the detection and feature extraction of underground targets. A Disentanglement Representation Generative Adversarial network (DR-GAN) for clutter suppression in GPR B-scan images is proposed. A target feature encoder and a clutter feature encoder are designed to extract target features and clutter features in GPR B-scan images. A clutter suppression generator is designed to obtain the GPR B-scan image after clutter suppression. Compared with the existing GPR clutter suppression methods based on supervised learning, the proposed method does not need pairwise matching data during network training, and can be better applied to the clutter suppression of measured GPR images. Experimental results on simulated and measured GPR data show that DR-GAN is an unsupervised learning network with better clutter suppression performance. The data of reinforcement embedded in quartz sand are collected, and the measured data containing clutter are processed by DR-GAN. The Improvement Factor (IF) index of the processing results is 17.85 dB higher than that of the existing Robust Nonnegative Matrix Factorization (RNMF) method.
Ground Penetrating Radar (GPR) is an underground nondestructive detection technology based on electromagnetic wave, which is widely used in municipal engineering, transportation, military and other fields. In the process of data acquisition, due to the coupling between transmitting antenna and receiving antenna, scattering from undulating ground and the complexity of underground random media, there is usually clutter in the GPR B-scan, which affects seriously the detection and feature extraction of underground targets. A Disentanglement Representation Generative Adversarial network (DR-GAN) for clutter suppression in GPR B-scan images is proposed. A target feature encoder and a clutter feature encoder are designed to extract target features and clutter features in GPR B-scan images. A clutter suppression generator is designed to obtain the GPR B-scan image after clutter suppression. Compared with the existing GPR clutter suppression methods based on supervised learning, the proposed method does not need pairwise matching data during network training, and can be better applied to the clutter suppression of measured GPR images. Experimental results on simulated and measured GPR data show that DR-GAN is an unsupervised learning network with better clutter suppression performance. The data of reinforcement embedded in quartz sand are collected, and the measured data containing clutter are processed by DR-GAN. The Improvement Factor (IF) index of the processing results is 17.85 dB higher than that of the existing Robust Nonnegative Matrix Factorization (RNMF) method.
2023, 45(10): 3786-3795.
doi: 10.11999/JEIT220997
Abstract:
To solve the problem of extended target tracking with non-star-convex irregular shape in complex environments, a level-set gaussian process extended target tracking algorithm based on energy functional is proposed. First, the interior of the shape is modeled by the polygonal method using the Level-Set Random Hypersurface Model (Level-Set RHM). Then, the nonlinear mapping relationship between the input and output of the Level-Set modeling is learned by using Gaussian Process (GP) to obtain the maximum value of the boundary function, and the nonlinear measurement equation based on the fusion of Level-Set and GP is further derived. Under the framework of optimal nonlinear filtering, Level-Set Gaussian Process (Level-Set GP) non-star convex extended target tracking algorithm is finally derived. And the area error is used as an evaluation index for the shape estimation of irregularly shaped extended targets. The simulation experiments show that the proposed algorithm is effective for the non-star convex irregular shape extended target shape estimation.
To solve the problem of extended target tracking with non-star-convex irregular shape in complex environments, a level-set gaussian process extended target tracking algorithm based on energy functional is proposed. First, the interior of the shape is modeled by the polygonal method using the Level-Set Random Hypersurface Model (Level-Set RHM). Then, the nonlinear mapping relationship between the input and output of the Level-Set modeling is learned by using Gaussian Process (GP) to obtain the maximum value of the boundary function, and the nonlinear measurement equation based on the fusion of Level-Set and GP is further derived. Under the framework of optimal nonlinear filtering, Level-Set Gaussian Process (Level-Set GP) non-star convex extended target tracking algorithm is finally derived. And the area error is used as an evaluation index for the shape estimation of irregularly shaped extended targets. The simulation experiments show that the proposed algorithm is effective for the non-star convex irregular shape extended target shape estimation.
2023, 45(10): 3796-3804.
doi: 10.11999/JEIT221265
Abstract:
The speed and frequency of underwater sound wave are much lower compared with the electromagnetic wave in the air. The moving target detection performance of multiple Unmanned Underwater Vehicle(multi-UUV) cooperative detection sonar is seriously influenced by the Doppler effect and narrowband signal processing. In this paper, the Discrete Frequency Coding Waveform of Non-Orthogonal Frequency Division Linear Frequency Modulation(DFCW-NOFD-LFM) based on Costas sequence and Orthogonal Frequency Division Linear Frequency Modulation(OFD-LFM) waveform is designed for multi-UUV cooperative detection sonar. The performance comparisons between the designed waveform and the traditional Code Division Multiple Access(CDMA) waveforms such as Binary Phase Shift Keying(BPSK), Discrete Frequency Coding Waveform(DFCW) are carried out. The results show that the designed DFCW-NOFD-LFM has large Doppler tolerance, good performance of auto-correlation, cross-correlation and reverberation suppression. It may be applied to the multi-UUV cooperative detection sonar to improve the relative moving target detection performance.
The speed and frequency of underwater sound wave are much lower compared with the electromagnetic wave in the air. The moving target detection performance of multiple Unmanned Underwater Vehicle(multi-UUV) cooperative detection sonar is seriously influenced by the Doppler effect and narrowband signal processing. In this paper, the Discrete Frequency Coding Waveform of Non-Orthogonal Frequency Division Linear Frequency Modulation(DFCW-NOFD-LFM) based on Costas sequence and Orthogonal Frequency Division Linear Frequency Modulation(OFD-LFM) waveform is designed for multi-UUV cooperative detection sonar. The performance comparisons between the designed waveform and the traditional Code Division Multiple Access(CDMA) waveforms such as Binary Phase Shift Keying(BPSK), Discrete Frequency Coding Waveform(DFCW) are carried out. The results show that the designed DFCW-NOFD-LFM has large Doppler tolerance, good performance of auto-correlation, cross-correlation and reverberation suppression. It may be applied to the multi-UUV cooperative detection sonar to improve the relative moving target detection performance.
2023, 45(10): 3805-3811.
doi: 10.11999/JEIT221061
Abstract:
According to the problems that the mismatch between the real source location and the dictionary grid leads to the great error of Direction Of Arrival (DOA), an Off-Grid DOA estimation based on modified Iterative Adaptive Approach (IAA) is proposed (OGIAA). Firstly, the signal power spectrum obtained by the IAA is modified, and the corresponding grid angle of the peak power is read out as the coarse estimation results. Then, the square error cost function is expanded to the second order Taylor expansion and minimized to obtain the initial offset. Finally, the power component and offset are optimized alternately to achieve high precision off grid DOA estimation. Theoretical analysis and simulation results show that the implementation process of this method is simple, and it can accurately estimate the source angle of the migration grid without regularization parameters. It also has higher estimation accuracy on non-uniform arrays with more degrees of freedom.
According to the problems that the mismatch between the real source location and the dictionary grid leads to the great error of Direction Of Arrival (DOA), an Off-Grid DOA estimation based on modified Iterative Adaptive Approach (IAA) is proposed (OGIAA). Firstly, the signal power spectrum obtained by the IAA is modified, and the corresponding grid angle of the peak power is read out as the coarse estimation results. Then, the square error cost function is expanded to the second order Taylor expansion and minimized to obtain the initial offset. Finally, the power component and offset are optimized alternately to achieve high precision off grid DOA estimation. Theoretical analysis and simulation results show that the implementation process of this method is simple, and it can accurately estimate the source angle of the migration grid without regularization parameters. It also has higher estimation accuracy on non-uniform arrays with more degrees of freedom.
2023, 45(10): 3812-3820.
doi: 10.11999/JEIT221435
Abstract:
The current design of the MetaSurface Absorber (MSA) has excellent absorption characteristics of ElectroMagnetic (EM) wave. However, the influence of incident EM wavefrom and power is rarely considered. In this paper, a nonlinear circuit based MSA is proposed, which can absorb selectively wave with specific waveforms at the same frequency. The unit-cell structure of the proposed MSA is composed of a metal square ring resonator integrated with a nonlinear circuit of four diodes parallel with one resistor and capacitor, an intermediate dielectric substrate (Rogers4350) isolation layer and a bottom ground plane. The results show that the designed MSA can achieve an absorbance of 97% of the 50 ns short pulse at 3.2 GHz, while the corresponding one of the continuous wave is only 21% at the same frequency. Around 3.2 GHz, the absorbance of 50 ns pulse can be dynamically adjusted by changing incident pulse power, and the absorbance is always above 60%, while the corresponding one of the continuous wave is only fixed at about 20%. When the pulse width of the incident wave is increased, the absorption level of the designed MSA firstly increases and then decreases significantly. In the case of oblique incident wave of TE and TM modes with 0 dBm power and −4 dBm, the absorbance of the designed MSA is still more than 60%. Further research results show that the absorption characteristics of the MSA is depend heavily on the design of capacitance as well as the geometric parameters of the unit-cell structure. This research has potential application prospects to wireless communication, anti-EM interference, EM compatibility and other fields.
The current design of the MetaSurface Absorber (MSA) has excellent absorption characteristics of ElectroMagnetic (EM) wave. However, the influence of incident EM wavefrom and power is rarely considered. In this paper, a nonlinear circuit based MSA is proposed, which can absorb selectively wave with specific waveforms at the same frequency. The unit-cell structure of the proposed MSA is composed of a metal square ring resonator integrated with a nonlinear circuit of four diodes parallel with one resistor and capacitor, an intermediate dielectric substrate (Rogers4350) isolation layer and a bottom ground plane. The results show that the designed MSA can achieve an absorbance of 97% of the 50 ns short pulse at 3.2 GHz, while the corresponding one of the continuous wave is only 21% at the same frequency. Around 3.2 GHz, the absorbance of 50 ns pulse can be dynamically adjusted by changing incident pulse power, and the absorbance is always above 60%, while the corresponding one of the continuous wave is only fixed at about 20%. When the pulse width of the incident wave is increased, the absorption level of the designed MSA firstly increases and then decreases significantly. In the case of oblique incident wave of TE and TM modes with 0 dBm power and −4 dBm, the absorbance of the designed MSA is still more than 60%. Further research results show that the absorption characteristics of the MSA is depend heavily on the design of capacitance as well as the geometric parameters of the unit-cell structure. This research has potential application prospects to wireless communication, anti-EM interference, EM compatibility and other fields.
