Submit Manuscript
Peer Review
Editor Work2019 Vol. 8, No. 4
Previous Issue
2019, 8(4): 425-435.
Abstract
1976KB
Due to the existing spectral clustering methods have low accuracy for PolSAR image classification, a Markov-based Discriminative Spectral Clustering(MDSC) method is proposed, which has the characteristics of low-rank and sparse decomposition. Firstly, a real low-rank probability transfer matrix is restored as an input to the standard Markov spectral clustering method to reduce the influence of noise on the classification result. Then the discriminative information is introduced into the objective function to make the polarimetric SAR image data can be more fully used. Finally, the augmented Lagrangian multiplier method is used to solve the objective function optimization problem under low-rank and probability simplex constraints. Experiments on three different data sets of Flevoland, Oberpfaffenhofen, and Xi’an show that our method has good accuracy and low sensitivity, which having a good classification performance.
Due to the existing spectral clustering methods have low accuracy for PolSAR image classification, a Markov-based Discriminative Spectral Clustering(MDSC) method is proposed, which has the characteristics of low-rank and sparse decomposition. Firstly, a real low-rank probability transfer matrix is restored as an input to the standard Markov spectral clustering method to reduce the influence of noise on the classification result. Then the discriminative information is introduced into the objective function to make the polarimetric SAR image data can be more fully used. Finally, the augmented Lagrangian multiplier method is used to solve the objective function optimization problem under low-rank and probability simplex constraints. Experiments on three different data sets of Flevoland, Oberpfaffenhofen, and Xi’an show that our method has good accuracy and low sensitivity, which having a good classification performance.
2019, 8(4): 436-447.
To overcome the difficulty of similarity expression and the effects of speckle noise in unsupervised classification of Polarimetric Synthetic Aperture Radar (PolSAR) images, a novel unsupervised PolSAR image terrain classification algorithm based on Tensor Product Graph (TPG) diffusion has been developed herein. Generally, TPG diffusion is usually utilized for optical image segmentation or image retrieval. In the present study, it can be used for PolSAR image terrain classification. TPG diffusion can robustly estimate geodesic distances ; therefore, it can be used for mining the intrinsic affinity between data points. First, the PolSAR image is over-segmented into many superpixels. Second, seven features are extracted based on the segmented superpixels to form a feature vector and construct a similarity matrix by using the Gaussian kernel. Third, TPG diffusion is performed on this similarity matrix to obtain another similarity matrix with stronger discriminability by propagating affinity information along the mainfold structure of data to achieve the global affinity measure. Finally, spectral clustering based on the diffused similarity matrix is adopted to perform terrain classification. Extensive experiments conducted on both simulated and real-world PolSAR images demonstrate that our approach can effectively combine neighborhood information and achieve higher classification accuracy, compared to four other competitive state-of-the-art methods.
To overcome the difficulty of similarity expression and the effects of speckle noise in unsupervised classification of Polarimetric Synthetic Aperture Radar (PolSAR) images, a novel unsupervised PolSAR image terrain classification algorithm based on Tensor Product Graph (TPG) diffusion has been developed herein. Generally, TPG diffusion is usually utilized for optical image segmentation or image retrieval. In the present study, it can be used for PolSAR image terrain classification. TPG diffusion can robustly estimate geodesic distances ; therefore, it can be used for mining the intrinsic affinity between data points. First, the PolSAR image is over-segmented into many superpixels. Second, seven features are extracted based on the segmented superpixels to form a feature vector and construct a similarity matrix by using the Gaussian kernel. Third, TPG diffusion is performed on this similarity matrix to obtain another similarity matrix with stronger discriminability by propagating affinity information along the mainfold structure of data to achieve the global affinity measure. Finally, spectral clustering based on the diffused similarity matrix is adopted to perform terrain classification. Extensive experiments conducted on both simulated and real-world PolSAR images demonstrate that our approach can effectively combine neighborhood information and achieve higher classification accuracy, compared to four other competitive state-of-the-art methods.
2019, 8(4): 448-457.
Given the problems that the amount of supervised information in the Polarimetric Synthetic Aperture Radar (PolSAR) image is low and the speckle noise is difficult to eliminate, in this study, a robust classification algorithm for PolSAR image based on Pinball loss Support Vector Machine (Pin-SVM) is proposed from the perspective of robust statistical learning. On the basis of the scattering characteristics of PolSAR images and the texture characteristics of surface features, the proposed algorithm determines the optimal decision hyperplane by solving the maximum quantile distance between the samples of two classes, which can provide more robust results without iteration. Compared with the traditional PolSAR image classification algorithms that solve the maximum margin, on one hand, the proposed algorithm is robust to the noise contained in the features extracted from PolSAR images. On the other hand, the proposed algorithm is insensitive to the sampling range of training samples, which means that it has better robustness to resampling. The experimental results of EMISAR-Foulum PolSAR data prove the validity of the proposed algorithm through comparative tests in a variety of situations.
Given the problems that the amount of supervised information in the Polarimetric Synthetic Aperture Radar (PolSAR) image is low and the speckle noise is difficult to eliminate, in this study, a robust classification algorithm for PolSAR image based on Pinball loss Support Vector Machine (Pin-SVM) is proposed from the perspective of robust statistical learning. On the basis of the scattering characteristics of PolSAR images and the texture characteristics of surface features, the proposed algorithm determines the optimal decision hyperplane by solving the maximum quantile distance between the samples of two classes, which can provide more robust results without iteration. Compared with the traditional PolSAR image classification algorithms that solve the maximum margin, on one hand, the proposed algorithm is robust to the noise contained in the features extracted from PolSAR images. On the other hand, the proposed algorithm is insensitive to the sampling range of training samples, which means that it has better robustness to resampling. The experimental results of EMISAR-Foulum PolSAR data prove the validity of the proposed algorithm through comparative tests in a variety of situations.
2019, 8(4): 458-470.
In this paper, a novel semi-supervised classification method based on the Neighborhood Minimum Spanning Tree (NMST) is proposed to solve the Polarimetric Synthetic Aperture Radar (PolSAR) terrain classification when labeled samples are few. Combining the idea of self-training method and spatial information of the pixels in PolSAR image, a new help-training sample selection strategy based on spatial neighborhood information is proposed, named as NMST, to select the high reliable unlabeled samples to enlarge the training set and improve the base classifier. Finally, the PolSAR image is classified by this improved classifier. The experiments results tested on three PolSAR data sets show that the proposed method achieves a better performance than existing classification methods when the number of labeled samples is few.
In this paper, a novel semi-supervised classification method based on the Neighborhood Minimum Spanning Tree (NMST) is proposed to solve the Polarimetric Synthetic Aperture Radar (PolSAR) terrain classification when labeled samples are few. Combining the idea of self-training method and spatial information of the pixels in PolSAR image, a new help-training sample selection strategy based on spatial neighborhood information is proposed, named as NMST, to select the high reliable unlabeled samples to enlarge the training set and improve the base classifier. Finally, the PolSAR image is classified by this improved classifier. The experiments results tested on three PolSAR data sets show that the proposed method achieves a better performance than existing classification methods when the number of labeled samples is few.
2019, 8(4): 471-478.
In recent years, Polarimetric Synthetic Aperture Radar (PolSAR) image classification has been investigated extensively. The traditional PolSAR image terrain classification methods result in a weak feature representation. To overcome this limitation, this study aims to propose a terrain classification method based on deep Convolutional Neural Network (CNN) and Conditional Random Field (CRF). The pre-trained VGG-Net-16 model was used to extract more powerful image features, and then the terrain from the images was classified through the efficient use of multiple features and context information by conditional random fields. The experimental results show that the proposed method can extract more features effectively in comparison with the three methods using traditional classical features and it can also achieve a higher overall accuracy and Kappa coefficient.
In recent years, Polarimetric Synthetic Aperture Radar (PolSAR) image classification has been investigated extensively. The traditional PolSAR image terrain classification methods result in a weak feature representation. To overcome this limitation, this study aims to propose a terrain classification method based on deep Convolutional Neural Network (CNN) and Conditional Random Field (CRF). The pre-trained VGG-Net-16 model was used to extract more powerful image features, and then the terrain from the images was classified through the efficient use of multiple features and context information by conditional random fields. The experimental results show that the proposed method can extract more features effectively in comparison with the three methods using traditional classical features and it can also achieve a higher overall accuracy and Kappa coefficient.
2019, 8(4): 479-489.
In the study of the terrain classification based on the Polarimetric Synthetic Aperture Radar (PolSAR), the algorithms based on general CNN do not fully utilize the phase information in different channels, and the pixel-by-pixel classification strategy with extensive redundant computation is inefficient. To mitigate these problems, a deep pixel-to-pixel mapping model in the complex domain is used for achieving a fast and accurate PolSAR terrain classification at a low sampling rate. To completely utilize the phase information, this study uses Group-Cross CNN to extend the original model to the complex domain allowing complex-number input signals and significantly improving the classification accuracy. In addition, to speed up the algorithm, a Fine-tuned Dilated Group-Cross CNN (FDGC-CNN) was adopted to directly achieve pixel-to-pixel mapping as well as improve accuracy. We verified the adopted model on two PolSAR images comprising 16 classes terrains from the AIRSAR and 4 classes terrains from the E-SAR. According to our model, the overall classification accuracies were 96.94% and 90.07% respectively while the running time was 4.22 s and 4.02 s respectively. Therefore, FDGC-CNN achieved better accuracy with higher efficiency compared to SVM and traditional CNN.
In the study of the terrain classification based on the Polarimetric Synthetic Aperture Radar (PolSAR), the algorithms based on general CNN do not fully utilize the phase information in different channels, and the pixel-by-pixel classification strategy with extensive redundant computation is inefficient. To mitigate these problems, a deep pixel-to-pixel mapping model in the complex domain is used for achieving a fast and accurate PolSAR terrain classification at a low sampling rate. To completely utilize the phase information, this study uses Group-Cross CNN to extend the original model to the complex domain allowing complex-number input signals and significantly improving the classification accuracy. In addition, to speed up the algorithm, a Fine-tuned Dilated Group-Cross CNN (FDGC-CNN) was adopted to directly achieve pixel-to-pixel mapping as well as improve accuracy. We verified the adopted model on two PolSAR images comprising 16 classes terrains from the AIRSAR and 4 classes terrains from the E-SAR. According to our model, the overall classification accuracies were 96.94% and 90.07% respectively while the running time was 4.22 s and 4.02 s respectively. Therefore, FDGC-CNN achieved better accuracy with higher efficiency compared to SVM and traditional CNN.
2019, 8(4): 490-500.
Recently, Netted Radar System (NRS) has received much attention due to its robust performance gain. Usually, the NRS Detect Before Track (DBT) method detects the received data at each time, acquiring a set of alarm plots, and then transmits these plots or the trajectories obtained based on them to the fusion center, thus generating a global estimated result. However, when the Signal-to-Noise Ratio (SNR) is low, the performance becomes highly degraded because the targets cannot pass the single-frame detection threshold of DBT. To solve this problem, a netted radar Multi-Frame Track Before Detect (MF-TBD) method is proposed in this paper. First, MF-TBD is performed in local radar nodes, and then it acquires estimated target plot sequences and transmits them to the center for further fusion. MF-TBD can take advantage of NRS, and also can utilize target space time correlation through MF-TBD processing and enhance the target SNR. Thus, it can improve detection performance. However, the outputs of MF-TBD are different from that of DBT. Therefore, the current fusion methods for DBT are not suitable for MF-TBD. To solve this problem, this paper first derives a fusion method for plot sequences, then reports its processing steps in radar system, and finally proposes an implementation method based on the particle filter. The simulation results show that the proposed method has a detection performance gain of 4 to 6 dB than the traditional method based on DBT, and a 50% gain on estimation accuracy than single-sensor MF-TBD.
Recently, Netted Radar System (NRS) has received much attention due to its robust performance gain. Usually, the NRS Detect Before Track (DBT) method detects the received data at each time, acquiring a set of alarm plots, and then transmits these plots or the trajectories obtained based on them to the fusion center, thus generating a global estimated result. However, when the Signal-to-Noise Ratio (SNR) is low, the performance becomes highly degraded because the targets cannot pass the single-frame detection threshold of DBT. To solve this problem, a netted radar Multi-Frame Track Before Detect (MF-TBD) method is proposed in this paper. First, MF-TBD is performed in local radar nodes, and then it acquires estimated target plot sequences and transmits them to the center for further fusion. MF-TBD can take advantage of NRS, and also can utilize target space time correlation through MF-TBD processing and enhance the target SNR. Thus, it can improve detection performance. However, the outputs of MF-TBD are different from that of DBT. Therefore, the current fusion methods for DBT are not suitable for MF-TBD. To solve this problem, this paper first derives a fusion method for plot sequences, then reports its processing steps in radar system, and finally proposes an implementation method based on the particle filter. The simulation results show that the proposed method has a detection performance gain of 4 to 6 dB than the traditional method based on DBT, and a 50% gain on estimation accuracy than single-sensor MF-TBD.
2019, 8(4): 501-509.
For collocated Multiple-Input Multiple-Output (MIMO) radar, we investigate the target detection problem in Gaussian clutter with an unknown but random covariance matrix. An inverse complex Wishart distribution is chosen as prior knowledge for the random covariance matrix. We propose two detectors in the Bayesian framework based on the criteria of the Generalized Likelihood Ratio Test. The two main advantages of the proposed Bayesian detectors are as follows: (1) no training data are required; and (2) a prior knowledge about the clutter is incorporated in the decision rules to achieve detection performance gains. Numerical simulations show that the proposed Bayesian detectors outperform the current commonly used non-Bayesian counterparts, particularly when the sample number of the transmitted waveform is small. In addition, the performance of the proposed detector will decline in parameter mismatched situation.
For collocated Multiple-Input Multiple-Output (MIMO) radar, we investigate the target detection problem in Gaussian clutter with an unknown but random covariance matrix. An inverse complex Wishart distribution is chosen as prior knowledge for the random covariance matrix. We propose two detectors in the Bayesian framework based on the criteria of the Generalized Likelihood Ratio Test. The two main advantages of the proposed Bayesian detectors are as follows: (1) no training data are required; and (2) a prior knowledge about the clutter is incorporated in the decision rules to achieve detection performance gains. Numerical simulations show that the proposed Bayesian detectors outperform the current commonly used non-Bayesian counterparts, particularly when the sample number of the transmitted waveform is small. In addition, the performance of the proposed detector will decline in parameter mismatched situation.
2019, 8(4): 510-518.
This study investigates the influence of the target polarization scattering characteristics on the detection performance of Polarization Diversity Radar (PDR) system with three typical working modes: full polarization mode, mixed polarization mode, and single polarization mode. The statistical properties of target echo signal corresponding to the three modes are separately deduced based on the statistical model of target polarization scattering and the receiving voltage equations. Furthermore, the optimal polarization diversity multi-channel fusion detection algorithm is designed under the Neyman-Pearson criterion in the background of Gaussian distribution, and closed-form expressions of probability of false alarm and detection are derived. The simulation results show that the correlations between the polarization scattering components of the target, particularly the correlations between the matched and cross polarization scattering components of the target, have greater effect on the detection performance of PDR for a given system Signal-to-Noise Ratio (SNR). In addition, the detection performance of the full and single polarization modes is more robust than that of the mixed polarization mode.
This study investigates the influence of the target polarization scattering characteristics on the detection performance of Polarization Diversity Radar (PDR) system with three typical working modes: full polarization mode, mixed polarization mode, and single polarization mode. The statistical properties of target echo signal corresponding to the three modes are separately deduced based on the statistical model of target polarization scattering and the receiving voltage equations. Furthermore, the optimal polarization diversity multi-channel fusion detection algorithm is designed under the Neyman-Pearson criterion in the background of Gaussian distribution, and closed-form expressions of probability of false alarm and detection are derived. The simulation results show that the correlations between the polarization scattering components of the target, particularly the correlations between the matched and cross polarization scattering components of the target, have greater effect on the detection performance of PDR for a given system Signal-to-Noise Ratio (SNR). In addition, the detection performance of the full and single polarization modes is more robust than that of the mixed polarization mode.
2019, 8(4): 519-526.
Ground Penetrating Radar (GPR) is a widely used non-destructive testing tool. Constructing an appropriate forward model is crucial for GPR to perform a full-waveform inversion of layered media. In this paper, a forward model for the quasi-monostatic Stepped-Frequency GPR (SFGPR) is proposed. In the model, the GPR and its interaction with the layered medium are represented as a linear equation in which the effects of the antennas are represented by a set of frequency-dependent transfer functions. To verify the accuracy of the proposed model, the authors constructed a quasi-monostatic SFGPR in a laboratory condition and performed a full-waveform inversion of the measurement signals of plasterboard and woodblock with known thickness. In the inversion results, the thickness estimation errors of the plasterboard and woodblock are not more than 0.3 mm, indicating that the proposed forward model has a very high accuracy. The inversion performances of the quasi-monostatic and monostatic SFGPR are further compared for the layered medium constructed with plasterboard and woodblock, which has a small permittivity difference. The results show that the quasi-monostatic SFGPR can obtain more accurate inversion parameters. By estimating the Signal to Noise Ratio (SNR) of the reflected signal from the interface, it is found that the SNR obtained by the quasi-monostatic configuration is 10 dB higher than that of the monostatic; therefore, the quasi-monostatic GPR has the better inversion performance.
Ground Penetrating Radar (GPR) is a widely used non-destructive testing tool. Constructing an appropriate forward model is crucial for GPR to perform a full-waveform inversion of layered media. In this paper, a forward model for the quasi-monostatic Stepped-Frequency GPR (SFGPR) is proposed. In the model, the GPR and its interaction with the layered medium are represented as a linear equation in which the effects of the antennas are represented by a set of frequency-dependent transfer functions. To verify the accuracy of the proposed model, the authors constructed a quasi-monostatic SFGPR in a laboratory condition and performed a full-waveform inversion of the measurement signals of plasterboard and woodblock with known thickness. In the inversion results, the thickness estimation errors of the plasterboard and woodblock are not more than 0.3 mm, indicating that the proposed forward model has a very high accuracy. The inversion performances of the quasi-monostatic and monostatic SFGPR are further compared for the layered medium constructed with plasterboard and woodblock, which has a small permittivity difference. The results show that the quasi-monostatic SFGPR can obtain more accurate inversion parameters. By estimating the Signal to Noise Ratio (SNR) of the reflected signal from the interface, it is found that the SNR obtained by the quasi-monostatic configuration is 10 dB higher than that of the monostatic; therefore, the quasi-monostatic GPR has the better inversion performance.
2019, 8(4): 527-536.
This paper focuses on an improved imaging-algorithm for the spotlight Synthetic Aperture Radar (spotlight SAR) with continuous Pulse Repetition Frequency (PRF) variation in extremely high-resolution imaging-process. PRI variation is conventionally employed to resolve the problem of fixed blind ranges as well as the conflict of high-resolution and wide-swath; however, there are problems such as spectrum aliasing and ambiguous targets caused by nonuniform sampling. In this study, a novel sinc interpolation method is proposed to reconstruct a uniformly sampled signal from non-uniform Fourier Transform samples. Then a two-step processing approach combined with the novel sinc interpolation method is presented in the process of non-uniformly sampled echo imaging. The simulation proves the validity and accuracy of the proposed imaging algorithm. In addition, the computational cost of the novel sinc interpolation is further reduced compared to that of non-uniform Fourier transformation.
This paper focuses on an improved imaging-algorithm for the spotlight Synthetic Aperture Radar (spotlight SAR) with continuous Pulse Repetition Frequency (PRF) variation in extremely high-resolution imaging-process. PRI variation is conventionally employed to resolve the problem of fixed blind ranges as well as the conflict of high-resolution and wide-swath; however, there are problems such as spectrum aliasing and ambiguous targets caused by nonuniform sampling. In this study, a novel sinc interpolation method is proposed to reconstruct a uniformly sampled signal from non-uniform Fourier Transform samples. Then a two-step processing approach combined with the novel sinc interpolation method is presented in the process of non-uniformly sampled echo imaging. The simulation proves the validity and accuracy of the proposed imaging algorithm. In addition, the computational cost of the novel sinc interpolation is further reduced compared to that of non-uniform Fourier transformation.
微信 | 公众平台 
