In recent years,deeps learning has been widely applied in synthetic aperture radar(SAR)image processing.However,the collection of large-scale labeled SAR images is challenging and costly,and the classification accurac...In recent years,deeps learning has been widely applied in synthetic aperture radar(SAR)image processing.However,the collection of large-scale labeled SAR images is challenging and costly,and the classification accuracy is often poor when only limited SAR images are available.To address this issue,we propose a novel framework for sparse SAR target classification under few-shot cases,termed the transfer learning-based interpretable lightweight convolutional neural network(TL-IL-CNN).Additionally,we employ enhanced gradient-weighted class activation mapping(Grad-CAM)to mitigate the“black box”effect often associated with deep learning models and to explore the mechanisms by which a CNN classifies various sparse SAR targets.Initially,we apply a novel bidirectional iterative soft thresholding(BiIST)algorithm to generate sparse images of superior quality compared to those produced by traditional matched filtering(MF)techniques.Subsequently,we pretrain multiple shallow CNNs on a simulated SAR image dataset.Using the sparse SAR dataset as input for the CNNs,we assess the efficacy of transfer learning in sparse SAR target classification and suggest the integration of TL-IL-CNN to enhance the classification accuracy further.Finally,Grad-CAM is utilized to provide visual explanations for the predictions made by the classification framework.The experimental results on the MSTAR dataset reveal that the proposed TL-IL-CNN achieves nearly 90%classification accuracy with only 20%of the training data required under standard operating conditions(SOC),surpassing typical deep learning methods such as vision Transformer(ViT)in the context of small samples.Remarkably,it even presents better performance under extended operating conditions(EOC).Furthermore,the application of Grad-CAM elucidates the CNN’s differentiation process among various sparse SAR targets.The experiments indicate that the model focuses on the target and the background can differ among target classes.The study contributes to an enhanced understanding of the interpretability of such results and enables us to infer the classification outcomes for each category more accurately.展开更多
On the basis of scale invariant feature transform（SIFT） descriptors,a novel kind of local invariants based on SIFT sequence scale（SIFT-SS） is proposed and applied to target classification.First of all,the merits o...On the basis of scale invariant feature transform（SIFT） descriptors,a novel kind of local invariants based on SIFT sequence scale（SIFT-SS） is proposed and applied to target classification.First of all,the merits of using an SIFT algorithm for target classification are discussed.Secondly,the scales of SIFT descriptors are sorted by descending as SIFT-SS,which is sent to a support vector machine（SVM） with radial based function（RBF） kernel in order to train SVM classifier,which will be used for achieving target classification.Experimental results indicate that the SIFT-SS algorithm is efficient for target classification and can obtain a higher recognition rate than affine moment invariants（AMI） and multi-scale auto-convolution（MSA） in some complex situations,such as the situation with the existence of noises and occlusions.Moreover,the computational time of SIFT-SS is shorter than MSA and longer than AMI.展开更多
Passive target detection through shipping-radiated noise is a key technology in current underwater operations and is of great research value in civil and military fields.In this study,the stable spectral line componen...Passive target detection through shipping-radiated noise is a key technology in current underwater operations and is of great research value in civil and military fields.In this study,the stable spectral line component of shipping-radiated noise is used as the research object,and the classification of multisource targets is studied from the perspective of underwater channels.We utilize the channel impulse response function as the classification basis of different targets.First,the underwater channel is estimated by the cepstrum.Then,the channel cepstral features carried by different spectral line components are extracted in turn.Finally,the spectral line components belonging to the same target are clustered by the cepstral feature distance to realize the classification of different targets.The simulation and experimental results verify the effectiveness of the proposed method in this research.展开更多
The different approaches used for target decomposition （TD） theory in radar polarimetry are reviewed and three main types of theorems are introduced： those based on Mueller matrix, those using an eigenvector analys...The different approaches used for target decomposition （TD） theory in radar polarimetry are reviewed and three main types of theorems are introduced： those based on Mueller matrix, those using an eigenvector analysis of the coherency matrix, and those employing coherent decomposition of the scattering matrix. Support vector machine （SVM）, as a novel approach in pattern recognition, has demonstrated success in many fields. A new algorithm of target classification, by combining target decomposition and the support vector machine, is proposed. To conduct the experiment, the polarimetric synthetic aperture radar （SAR） data are used. Experimental results show that it is feasible and efficient to target classification by applying target decomposition to extract scattering mechanisms, and the effects of kernel function and its parameters on the classification efficiency are significant.展开更多
Surface/underwater target classification is a key topic in marine information research.However,the complex underwater environment,coupled with the diversity of target types and their variable characteristics,presents ...Surface/underwater target classification is a key topic in marine information research.However,the complex underwater environment,coupled with the diversity of target types and their variable characteristics,presents significant challenges for classifier design.For shallow-water waveguides with a negative thermocline,a residual neural network(ResNet)model based on the sound field elevation structure is constructed.This model demonstrates robust classification performance even when facing low signal-to-noise ratios and environmental mismatches.Meanwhile,to address the reduced generalization ability caused by limited labeled acoustic data,an improved ResNet model based on unsupervised domain adaptation(“proposed UDA-ResNet”)is further constructed.This model incorporates data on simulated elevation structures of the sound field to augment the training process.Adversarial training is employed to extract domain-invariant features from simulated and trial data.These strategies help reduce the negative impact caused by domain differences.Experimental results demonstrate that the proposed method shows strong surface/underwater target classification ability under limited sample sizes,thus confirming its feasibility and effectiveness.展开更多
A novel classification algorithm based on abnormal magnetic signals is proposed for ground moving targets which are made of ferromagnetic material. According to the effect of diverse targets on earth's magnetism,t...A novel classification algorithm based on abnormal magnetic signals is proposed for ground moving targets which are made of ferromagnetic material. According to the effect of diverse targets on earth's magnetism,the moving targets are detected by a magnetic sensor and classified with a simple computation method. The detection sensor is used for collecting a disturbance signal of earth magnetic field from an undetermined target. An optimum category match pattern of target signature is tested by training some statistical samples and designing a classification machine. Three ordinary targets are researched in the paper. The experimental results show that the algorithm has a low computation cost and a better sorting accuracy. This classification method can be applied to ground reconnaissance and target intrusion detection.展开更多
Imaging detection is an important means to obtain target information.The traditional imaging detection technology mainly collects the intensity information and spectral information of the target to realize the classif...Imaging detection is an important means to obtain target information.The traditional imaging detection technology mainly collects the intensity information and spectral information of the target to realize the classification of the target.In practical applications,due to the mixed scenario,it is difficult to meet the needs of target recognition.Compared with intensity detection,the method of polarization detection can effectively enhance the accuracy of ground object target recognition(such as the camouflage target).In this paper,the reflection mechanism of the target surface is studied from the microscopic point of view,and the polarization characteristic model is established to express the relationship between the polarization state of the reflected signal and the target surface parameters.The polarization characteristic test experiment is carried out,and the target surface parameters are retrieved using the experimental data.The results show that the degree of polarization(DOP)is closely related to the detection zenith angle and azimuth angle.The(DOP)of the target is the smallest in the direction of light source incidence and the largest in the direction of specular reflection.Different materials have different polarization characteristics.By comparing their DOP,target classification can be achieved.展开更多
In most of the passive tracking systems, only the target kinematical information is used in the measurement-to-track association, which results in error tracking in a multitarget environment, where the targets are too...In most of the passive tracking systems, only the target kinematical information is used in the measurement-to-track association, which results in error tracking in a multitarget environment, where the targets are too close to each other. To enhance the tracking accuracy, the target signal classification information （TSCI） should be used to improve the data association. The TSCI is integrated in the data association process using the JPDA （joint probabilistic data association）. The use of the TSCI in the data association can improve discrimination by yielding a purer track and preserving continuity. To verify the validity of the application of TSCI, two simulation experiments are done on an air target-tracing problem, that is, one using the TSCI and the other not using the TSCI. The final comparison shows that the use of the TSCI can effectively improve tracking accuracy.展开更多
With the rapidly growing abuse of drones, monitoring and classification of birds and drones have become a crucial safety issue. With similar low radar cross sections(RCSs), velocities, and heights, drones are usually ...With the rapidly growing abuse of drones, monitoring and classification of birds and drones have become a crucial safety issue. With similar low radar cross sections(RCSs), velocities, and heights, drones are usually difficult to be distinguished from birds in radar measurements. In this paper, we propose to exploit different periodical motions of birds and drones from highresolution Doppler spectrum sequences(DSSs) for classification.This paper presents an elaborate feature vector representing the periodic fluctuations of RCS and micro kinematics. Fed by the Doppler spectrum and feature sequence, the long to short-time memory(LSTM) is used to solve the time series classification.Different classification schemes to exploit the Doppler spectrum series are validated and compared by extensive real-data experiments, which confirms the effectiveness and superiorities of the proposed algorithm.展开更多
Based on pattern recognition theory and neural network technology, moving objects automatic detection and classification method integrating advanced wavelet analysis are discussed in detail. An algorithm of moving tar...Based on pattern recognition theory and neural network technology, moving objects automatic detection and classification method integrating advanced wavelet analysis are discussed in detail. An algorithm of moving targets pattern recognition on the combination of inter-frame difference and wavelet neural network is presented. The experimental results indicate that the designed BP wavelet network using this algorithm can recognize and classify moving targets rapidly and effectively.展开更多
This paper studies the generalization capability of feedforward neural networks (FNN).The mechanism of FNNs for classification is investigated from the geometric and probabilistic viewpoints. It is pointed out that th...This paper studies the generalization capability of feedforward neural networks (FNN).The mechanism of FNNs for classification is investigated from the geometric and probabilistic viewpoints. It is pointed out that the outputs of the output layer in the FNNs for classification correspond to the estimates of posteriori probability of the input pattern samples with desired outputs 1 or 0. The theorem for the generalized kernel function in the radial basis function networks (RBFN) is given. For an 2-layer perceptron network (2-LPN). an idea of using extended samples to improve generalization capability is proposed. Finally. the experimental results of radar target classification are given to verify the generaliztion capability of the RBFNs.展开更多
Optical remote sensing is a crucial component of the ocean observation system.However,the complex interactions between the ocean and atmosphere limit its observation capability and hinder the advancement of quantitati...Optical remote sensing is a crucial component of the ocean observation system.However,the complex interactions between the ocean and atmosphere limit its observation capability and hinder the advancement of quantitative applications and support capacity.Polarimetric remote sensing,as an advanced detection technology,investigates the anisotropic characteristics of electromagnetic waves perpendicular to the direction of propagation.Serving as an extension of conventional optical remote sensing,it significantly improves the accuracy of feature identification and quantitative estimation.As the most classical polarization feature,the Degree of Polarization(DoP)feature has been widely applied in various scenarios.In this study,the spatial distribution of the DoP feature over the 2πobservation space under oceanic conditions is thoroughly investigated through theoretical simulations and sample measurements.Our findings suggest that the DoP feature lacks sufficient sensitivity and versatility to be used independently in ocean observation scenarios.To address this limitation,a novel feature,namely the Angular Polarization(AP)feature,is proposed for polarimetric remote sensing tailored to ocean applications.The effectiveness of this new feature is validated in three representative ocean observation scenarios,and its performance is compared against both conventional optical feature and DoP feature.The results demonstrate that the AP feature offers distinct advantages in differentiating ocean bodies with varying refractive indices and in emphasizing the differences between observed targets.Moreover,its application enhances the accuracy of unsupervised classification for ocean observations.The establishment of the AP feature greatly strengthens the information-sensing capacity of polarimetric ocean remote sensing,offering a promising pathway to enhance the overall performance of ocean observation systems.展开更多
基金supported in part by the National Natural Science Foundation(Nos.62271248,62401256)in part by the Natural Science Foundation of Ji-angsu Province(Nos.BK20230090,BK20241384)in part by the Key Laboratory of Land Satellite Remote Sens-ing Application,Ministry of Natural Resources of China(No.KLSMNR-K202303)。
文摘In recent years,deeps learning has been widely applied in synthetic aperture radar(SAR)image processing.However,the collection of large-scale labeled SAR images is challenging and costly,and the classification accuracy is often poor when only limited SAR images are available.To address this issue,we propose a novel framework for sparse SAR target classification under few-shot cases,termed the transfer learning-based interpretable lightweight convolutional neural network(TL-IL-CNN).Additionally,we employ enhanced gradient-weighted class activation mapping(Grad-CAM)to mitigate the“black box”effect often associated with deep learning models and to explore the mechanisms by which a CNN classifies various sparse SAR targets.Initially,we apply a novel bidirectional iterative soft thresholding(BiIST)algorithm to generate sparse images of superior quality compared to those produced by traditional matched filtering(MF)techniques.Subsequently,we pretrain multiple shallow CNNs on a simulated SAR image dataset.Using the sparse SAR dataset as input for the CNNs,we assess the efficacy of transfer learning in sparse SAR target classification and suggest the integration of TL-IL-CNN to enhance the classification accuracy further.Finally,Grad-CAM is utilized to provide visual explanations for the predictions made by the classification framework.The experimental results on the MSTAR dataset reveal that the proposed TL-IL-CNN achieves nearly 90%classification accuracy with only 20%of the training data required under standard operating conditions(SOC),surpassing typical deep learning methods such as vision Transformer(ViT)in the context of small samples.Remarkably,it even presents better performance under extended operating conditions(EOC).Furthermore,the application of Grad-CAM elucidates the CNN’s differentiation process among various sparse SAR targets.The experiments indicate that the model focuses on the target and the background can differ among target classes.The study contributes to an enhanced understanding of the interpretability of such results and enables us to infer the classification outcomes for each category more accurately.
基金supported by the National High Technology Research and Development Program (863 Program) (2010AA7080302)
文摘On the basis of scale invariant feature transform（SIFT） descriptors,a novel kind of local invariants based on SIFT sequence scale（SIFT-SS） is proposed and applied to target classification.First of all,the merits of using an SIFT algorithm for target classification are discussed.Secondly,the scales of SIFT descriptors are sorted by descending as SIFT-SS,which is sent to a support vector machine（SVM） with radial based function（RBF） kernel in order to train SVM classifier,which will be used for achieving target classification.Experimental results indicate that the SIFT-SS algorithm is efficient for target classification and can obtain a higher recognition rate than affine moment invariants（AMI） and multi-scale auto-convolution（MSA） in some complex situations,such as the situation with the existence of noises and occlusions.Moreover,the computational time of SIFT-SS is shorter than MSA and longer than AMI.
基金This study was supported by the National Natural Sci-ence Foundation of China(No.11774073)the State Key Laboratory of Acoustics(No.SKLA201904).
文摘Passive target detection through shipping-radiated noise is a key technology in current underwater operations and is of great research value in civil and military fields.In this study,the stable spectral line component of shipping-radiated noise is used as the research object,and the classification of multisource targets is studied from the perspective of underwater channels.We utilize the channel impulse response function as the classification basis of different targets.First,the underwater channel is estimated by the cepstrum.Then,the channel cepstral features carried by different spectral line components are extracted in turn.Finally,the spectral line components belonging to the same target are clustered by the cepstral feature distance to realize the classification of different targets.The simulation and experimental results verify the effectiveness of the proposed method in this research.
文摘The different approaches used for target decomposition （TD） theory in radar polarimetry are reviewed and three main types of theorems are introduced： those based on Mueller matrix, those using an eigenvector analysis of the coherency matrix, and those employing coherent decomposition of the scattering matrix. Support vector machine （SVM）, as a novel approach in pattern recognition, has demonstrated success in many fields. A new algorithm of target classification, by combining target decomposition and the support vector machine, is proposed. To conduct the experiment, the polarimetric synthetic aperture radar （SAR） data are used. Experimental results show that it is feasible and efficient to target classification by applying target decomposition to extract scattering mechanisms, and the effects of kernel function and its parameters on the classification efficiency are significant.
基金supported by the National Natural Science Foundation of China(Grant Nos.62471024 and 62301183)the Open Research Fund of Hanjiang Laboratory(KF2024001).
文摘Surface/underwater target classification is a key topic in marine information research.However,the complex underwater environment,coupled with the diversity of target types and their variable characteristics,presents significant challenges for classifier design.For shallow-water waveguides with a negative thermocline,a residual neural network(ResNet)model based on the sound field elevation structure is constructed.This model demonstrates robust classification performance even when facing low signal-to-noise ratios and environmental mismatches.Meanwhile,to address the reduced generalization ability caused by limited labeled acoustic data,an improved ResNet model based on unsupervised domain adaptation(“proposed UDA-ResNet”)is further constructed.This model incorporates data on simulated elevation structures of the sound field to augment the training process.Adversarial training is employed to extract domain-invariant features from simulated and trial data.These strategies help reduce the negative impact caused by domain differences.Experimental results demonstrate that the proposed method shows strong surface/underwater target classification ability under limited sample sizes,thus confirming its feasibility and effectiveness.
基金Sponsored by the National Natural Science Foundation of China (60773129)the Excellent Youth Science and Technology Foundation of Anhui Province of China ( 08040106808)
文摘A novel classification algorithm based on abnormal magnetic signals is proposed for ground moving targets which are made of ferromagnetic material. According to the effect of diverse targets on earth's magnetism,the moving targets are detected by a magnetic sensor and classified with a simple computation method. The detection sensor is used for collecting a disturbance signal of earth magnetic field from an undetermined target. An optimum category match pattern of target signature is tested by training some statistical samples and designing a classification machine. Three ordinary targets are researched in the paper. The experimental results show that the algorithm has a low computation cost and a better sorting accuracy. This classification method can be applied to ground reconnaissance and target intrusion detection.
基金supported by the National Key Laboratory of Electromagnetic Space Security(JCKY2023230C009).
文摘Imaging detection is an important means to obtain target information.The traditional imaging detection technology mainly collects the intensity information and spectral information of the target to realize the classification of the target.In practical applications,due to the mixed scenario,it is difficult to meet the needs of target recognition.Compared with intensity detection,the method of polarization detection can effectively enhance the accuracy of ground object target recognition(such as the camouflage target).In this paper,the reflection mechanism of the target surface is studied from the microscopic point of view,and the polarization characteristic model is established to express the relationship between the polarization state of the reflected signal and the target surface parameters.The polarization characteristic test experiment is carried out,and the target surface parameters are retrieved using the experimental data.The results show that the degree of polarization(DOP)is closely related to the detection zenith angle and azimuth angle.The(DOP)of the target is the smallest in the direction of light source incidence and the largest in the direction of specular reflection.Different materials have different polarization characteristics.By comparing their DOP,target classification can be achieved.
基金the Youth Science and Technology Foundection of University of Electronic Science andTechnology of China (JX0622).
文摘In most of the passive tracking systems, only the target kinematical information is used in the measurement-to-track association, which results in error tracking in a multitarget environment, where the targets are too close to each other. To enhance the tracking accuracy, the target signal classification information （TSCI） should be used to improve the data association. The TSCI is integrated in the data association process using the JPDA （joint probabilistic data association）. The use of the TSCI in the data association can improve discrimination by yielding a purer track and preserving continuity. To verify the validity of the application of TSCI, two simulation experiments are done on an air target-tracing problem, that is, one using the TSCI and the other not using the TSCI. The final comparison shows that the use of the TSCI can effectively improve tracking accuracy.
基金supported by the National Natural Science Foundation of China (62101603)the Shenzhen Science and Technology Program(KQTD20190929172704911)+3 种基金the Aeronautical Science Foundation of China (2019200M1001)the National Nature Science Foundation of Guangdong (2021A1515011979)the Guangdong Key Laboratory of Advanced IntelliSense Technology (2019B121203006)the Pearl R iver Talent Recruitment Program (2019ZT08X751)。
文摘With the rapidly growing abuse of drones, monitoring and classification of birds and drones have become a crucial safety issue. With similar low radar cross sections(RCSs), velocities, and heights, drones are usually difficult to be distinguished from birds in radar measurements. In this paper, we propose to exploit different periodical motions of birds and drones from highresolution Doppler spectrum sequences(DSSs) for classification.This paper presents an elaborate feature vector representing the periodic fluctuations of RCS and micro kinematics. Fed by the Doppler spectrum and feature sequence, the long to short-time memory(LSTM) is used to solve the time series classification.Different classification schemes to exploit the Doppler spectrum series are validated and compared by extensive real-data experiments, which confirms the effectiveness and superiorities of the proposed algorithm.
文摘Based on pattern recognition theory and neural network technology, moving objects automatic detection and classification method integrating advanced wavelet analysis are discussed in detail. An algorithm of moving targets pattern recognition on the combination of inter-frame difference and wavelet neural network is presented. The experimental results indicate that the designed BP wavelet network using this algorithm can recognize and classify moving targets rapidly and effectively.
文摘This paper studies the generalization capability of feedforward neural networks (FNN).The mechanism of FNNs for classification is investigated from the geometric and probabilistic viewpoints. It is pointed out that the outputs of the output layer in the FNNs for classification correspond to the estimates of posteriori probability of the input pattern samples with desired outputs 1 or 0. The theorem for the generalized kernel function in the radial basis function networks (RBFN) is given. For an 2-layer perceptron network (2-LPN). an idea of using extended samples to improve generalization capability is proposed. Finally. the experimental results of radar target classification are given to verify the generaliztion capability of the RBFNs.
基金supported by the National Key R&D Program of China(Grant No.2023YFB3905703)the Strategic Priority Research Program of the Chinese Academy of Sciences(Grant No.XDA19090144)+1 种基金the National Natural Science Foundation of China(Grant No.41842048)the Shenzhen Academician(Expert)Workstation of Orbbec Inc.(XHXS2023-115)。
文摘Optical remote sensing is a crucial component of the ocean observation system.However,the complex interactions between the ocean and atmosphere limit its observation capability and hinder the advancement of quantitative applications and support capacity.Polarimetric remote sensing,as an advanced detection technology,investigates the anisotropic characteristics of electromagnetic waves perpendicular to the direction of propagation.Serving as an extension of conventional optical remote sensing,it significantly improves the accuracy of feature identification and quantitative estimation.As the most classical polarization feature,the Degree of Polarization(DoP)feature has been widely applied in various scenarios.In this study,the spatial distribution of the DoP feature over the 2πobservation space under oceanic conditions is thoroughly investigated through theoretical simulations and sample measurements.Our findings suggest that the DoP feature lacks sufficient sensitivity and versatility to be used independently in ocean observation scenarios.To address this limitation,a novel feature,namely the Angular Polarization(AP)feature,is proposed for polarimetric remote sensing tailored to ocean applications.The effectiveness of this new feature is validated in three representative ocean observation scenarios,and its performance is compared against both conventional optical feature and DoP feature.The results demonstrate that the AP feature offers distinct advantages in differentiating ocean bodies with varying refractive indices and in emphasizing the differences between observed targets.Moreover,its application enhances the accuracy of unsupervised classification for ocean observations.The establishment of the AP feature greatly strengthens the information-sensing capacity of polarimetric ocean remote sensing,offering a promising pathway to enhance the overall performance of ocean observation systems.
