Three-dimensional (3-D) geometry reconstruction and dynamic estimation of spinning spacecraft are crucial for space surveillance. Existing methods rely on interpreting 2-D inverse synthetic aperture radar (ISAR) images, which often suffer from image defocusing and scaling errors induced by target spin. To tackle this problem, we propose a joint 3-D ISAR imaging and dynamic estimation framework based on signal phase analysis, eliminating the need for prior 2-D imaging. After modeling the complex rotation of a spinning spacecraft relative to radar, we establish an overdetermined system of equations based on relationships between target parameters and cubic phase coefficients of radar echoes. Then, our two-step solution framework is organized as follows. First, we estimate cubic phase coefficients of the multicomponent signal through sparse representation (SR). To construct compact dictionaries for SR, we combine the coarse estimation and the phase order hierarchical processing (POHP) strategy to guide their design. Second, 3-D locations of scatterers, spin velocity, and direction of spin axis are estimated from the phase coefficients by solving the equation system. To ensure accurate and efficient solutions, we derive their iterative formulas using an alternating iterative strategy. With these formulas, 3-D ISAR imaging and dynamic estimation can be achieved directly. The effectiveness and superiority of our method are verified by experiments on simulated ISAR echoes.

In the field of synthetic aperture radar (SAR) vehicle detection, detectors based on deep learning have gained extensive application in recent years. However, most methods can only generate horizontal bounding boxes (HBBs). Due to the lack of angle information, the presence of excessive background noise in HBB leads to an inaccurate description of vehicles. Although research has been conducted on oriented SAR target detection, several issues persist, including the inadequate utilization of vehicle characteristics, inaccurate feature extraction for vehicles exhibiting size and rotation variations, and imprecise positioning of oriented bounding boxes (OBBs). To address these issues, we propose a novel-oriented SAR vehicle detection network. First, leveraging the shape property of vehicles, we devise a constrained rectangular-invariant rotatable convolution (Cons-RIRConv) for the backbone. Under the constraints of a carefully designed size- and rotation-invariant regularization term, Cons-RIRConv determines convolution sampling positions that are adaptive to the size and rotation variations of vehicles. This approach is conducive to Cons-RIRConv in extracting size- and rotation-invariant features of vehicles. Furthermore, as the sampling positions of Cons-RIRConv align closely with vehicles, we integrate them into the anchor generation process within the detection head, thereby developing an adaptive anchor generation mechanism (AAGM). Guided by the sampling positions of Cons-RIRConv, the anchor boxes generated by AAGM exhibit a high degree of conformity with the vehicles and low redundancy, ultimately improving the localization accuracy of OBB and reducing computation costs. Experiments on three authoritative measured SAR vehicle detection datasets show the effectiveness of our method.

Synthetic aperture radar (SAR) target recognition methods based on deep learning have been a research hot-spot. But most of them are of black-box structure and neglect physical characteristics of SAR targets, which restrict the recognition performance. Scattering centers (SCs) are physical features that reflect structure and shape information of targets. Thus, this paper designs a physics-driven interpretable SAR target recognition network based on scattering center feature extraction. Incorporating the scattering center model of SAR target into network, our method is interpretable deep model that learns the SC features with specific physical meanings. Moreover, the learned SC features are then constructed as the SC geometric images, which are further projected into a designed target recognition network for target recognition. In particular, our model is an end-to-end model that achieves SCs feature extraction and target recognition in a framework, which can avoid the mismatch issue between SC features and classifier and ensure promising recognition performance. Experiments on the measured MSTAR dataset validate the superior performance of our method.

Deep-learning-based methods have recently achieved significant advancements in synthetic aperture radar automatic target recognition (SAR ATR). However, these methods typically rely heavily on extensive annotations, which are difficult to obtain for SAR images. Semi-supervised learning offers a solution to improve model performance with limited labeled data by leveraging unlabeled data. The mainstream semi-supervised learning methods for SAR ATR typically select high-confidence unlabeled images to assign pseudo-labels for their inclusion in the model training process. However, the large number of low-confidence unlabeled images are not efficiently utilized. To address this issue, a semi-supervised SAR target recognition method based on contrastive learning and complementary label (CoL) learning is proposed. First, CoL learning assigns CoLto low-confidence unlabeled images based on their minimum prediction probabilities. Subsequently, a threshold is set to filter out unreliable CoL, thereby mitigating the adverse effects of erroneous CoL. This approach ensures the effective and comprehensive utilization of low-confidence unlabeled images. Additionally, we propose a contrastive loss that incorporates CoL. Compared to traditional contrastive losses, our proposed contrastive loss constructs a richer set of negative sample pairs by leveraging the characteristics of CoL more effectively. Consequently, this approach improves the utilization of low-confidence images and further improves recognition performance. In contrast to the current state-of-the-art semi-supervised recognition methods, experiments on the MSTAR dataset demonstrate the better recognition performance of our proposed method with limited labeled images.

Most existing synthetic aperture radar (SAR) target recognition methods based on deep learning are of black boxes structure and data-drive networks, which are faced with the issue of severe performance degradation under extended operating conditions (EOCs). To address the issue, this paper proposes an end-to-end mechanism-driven interpretable network (EMI-Net) for SAR target recognition under EOCs. The EMI-Net achieves the integration of scattering center feature extraction and target recognition in an end-to-end framework to avoid the mismatch between scattering center features with classifier and also explore the representative electromagnetic characteristics that are useful for recognition under EOCs. In EMI-Net, by unfolding a sparse solving algorithm and integrating scattering center model into deep networks, our model contains feature encoding and decoding procedures of SAR images. Thus, EMI-Net is an interpretable deep unfolding network that is driven by physical mechanism to precisely learn scattering center features reflecting locations and amplitude of SAR targets based on the deep learning mode. Our EMI-Net divides input images into multiple patches based on the image-domain scattering center model to reduce its computation and space complexities. In addition, EMI-Net treats the scattering centers as discrete three-dimensional point data and designs a point cloud network as classifier to explore permutation-invariant representations for recognition. Results on the measured dataset validate that EMI-Net gains superior scattering center extraction and target recognition results under EOCs, and also shows high time-efficiency and low memory requirement.

This article delves into the topic of unsupervised domain adaptation (UDA) by transferring knowledge from rich labeled optical domain to unlabeled synthetic aperture radar (SAR) domain, tackling the issues faced by deep-learning-based SAR ship classification methods that rely on abundant labeled SAR images. Typical UDA methods usually extract domain-invariant representations (DIRs) between two domains. However, due to the prominent differences in imaging mechanisms between optical and SAR images, the discriminative characteristics of same classes across domains may vary. Feature representation guided by labeled optical images therefore suffers from a particularly serious source-bias problem, making DIR difficult to be extracted. Moreover, capturing the category structure of the target domain is crucial for classification tasks. To solve the above challenges, this article proposes a UDA framework for SAR ship classification via progressive feature alignment between optical and unlabeled SAR domains, gradually aligning two domains across domain and class levels. At the domain level, to reduce the transfer difficulty stemming from the prominent differences between SAR and optical images, feature calibrated domain alignment (FCDA) is presented to achieve accurate DIR extraction. FCDA combines the reconstruction and the consistency constraints of different perturbed versions of the same image to calibrate the optical-bias representation into the features of unbiased toward a specific domain. At the class level, we proposed feature enhanced class alignment (FECA) to capture the fine-grained category structure of the SAR domain. FECA incorporates pseudo-label-based cross-domain contrastive learning (CDC) for intraclass compactness as well as interclass separation among cross-domain categories, along with a consistency learning approach to enhance the class structure of SAR domain. The experimental results indicate that our method achieves exceptional performance in unsupervised classification of SAR ships.

Radar-based gait recognition has gained wide attention recently for its ability to preserve privacy and adapt to low-light and poor-weather scenarios. Among different forms of input data, a radar point cloud is an appealing option as it captures not only the appearance signatures but also the motion signatures of the subject. For gait recognition, both appearance and motion are crucial signatures that can be represented by spatial features and temporal features, respectively. However, the spatial–temporal features extracted by existing radar point cloud-based methods are coarse-grained, leading to poor performance in realistic applications. To enhance the spatial–temporal feature representation ability of the radar point cloud-based method, in this article, we design a novel network to extract fine-grained spatial–temporal gait features from a millimeter-wave (mmWave) radar point cloud. For fine-grained spatial feature extraction, we apply a dual-stream feature extraction (DSFE) module to exploit the 3-D coordinates, intensity, and velocity information within the radar point cloud. After that, since each body part has its unique characteristics in the gait task, we also propose a probability-guided body-part partition (PGBP) module to split the radar point cloud into fine-grained spatial body parts. For fine-grained temporal feature extraction, a local–global temporal feature extraction (LGTE) module is used to further capture the temporal patterns of each body part. To evaluate the effectiveness of the proposed methods, we conduct extensive experiments with 90 subjects in various realistic settings, i.e., cross-view and cross-wearing conditions. The results demonstrate that our model achieves significant improvement over existing radar-based gait recognition methods.

In recent years, gait recognition has emerged as an important and promising solution for human identification. Generally, gait recognition is based on a single type of sensor, such as a camera or a radar. However, data of a single modality may only capture inadequate gait features of a person, such as camera data lacking the intuitive micro-motion pattern information and radar data lacking the information about gait appearance, making gait-based human identification system vulnerable to complex covariate conditions, e.g., cross-view and cross-walking-condition. To build a robust and reliable gait-based human identification system, in this study, we propose a multisensor gait recognition framework with deep convolutional neural networks (CNNs) by fusing camera gait energy images (GEIs) and radar time-Doppler spectrograms. To learn the fine-grained gait appearance features, we propose a body-part spatial attention (BPSA) module to obtain more discriminative body part representations of GEIs. To learn the gait micro-motion pattern, we propose a long-short temporal relation modeling (LSTRM) module to obtain the local and global micro-motion representation of time-Doppler spectrograms. Finally, we fuse the discriminative body part representation and the micro-motion pattern at the multiscale feature space to obtain richer and more robust gait features for human identification. We provide an extensive empirical evaluation in terms of various complex covariate conditions, namely, cross-view and cross-walking-condition. Experiments on 121 subjects with eight views and three walking conditions of camera and radar data show our proposed method is more robust and accurate.

The mainstream synthetic aperture radar (SAR) automatic target recognition (ATR) methods commonly use convolutional neural network (CNN) to extract the visual information of SAR targets, while the physical structural information is seldom considered. Scattering center features can describe the targets’ physical structure information and are robust to the local variations of targets, which can be exploited to reflect the local structural characteristic of SAR targets. Therefore, we propose a novel method that effectively combines global visual features and local structural features for SAR ATR. The local structural features here contain not only the local physical structure information but also the local visual information. Our proposed method consists of three parts: global-based module, local-based module, and feature fusion module. Global-based module utilizes CNN to extract global visual features from SAR images. Local-based module firstly extracts attributed scattering centers (ASCs) from the complex SAR image and models each ASC as a node to construct graph data, from which we further use a multi-scale graph convolutional network (GCN) to extract local structural features. The node features for GCN learning are constructed by multiplying the corresponding local visual features in shallow CNN feature maps with the ASC reconstruction maps to better reflect the local structure characteristic. Then the learned local structural features from GCN are further fused with global visual features to achieve SAR ATR. As far as authors know, this is the first work combining CNN and GCN to effectively extract global visual features and local structural features simultaneously in SAR ATR. Experiments on the Moving and Stationary Target Acquisition and Recognition (MSTAR) dataset show that our proposed method outperforms SOTA methods in terms of classification accuracy.

In recent years, target detection methods based on deep learning have achieved extensive development in synthetic aperture radar (SAR) ship detection. However, training such detectors requires target-level annotations of SAR images that are hard to be obtained in practice. To reduce the dependence of network training on expensive target-level annotations, we propose a novel semisupervised SAR ship detection network via scene characteristic learning. The proposed network focuses on utilizing the scene-level annotations of SAR images to improve the detection performance in the case of limited target-level annotations. Compared with the traditional fully supervised SAR ship detection network, the proposed network constructs a scene characteristic learning branch parallel with the detection branch. In the scene characteristic learning branch, a scene classification loss and a scene aggregation loss are designed to utilize the scene-level annotations. Under the constraint of these two losses, the feature extraction network can fully learn the scene characteristics of SAR images, thus enhancing its feature representation ability for ship targets and clutter. In addition, we propose a hierarchical test process from scene to target. After recognizing the scene types of input SAR images, we design different detection strategies for SAR images recognized as different scenes. The proposed test process can significantly reduce the inland and inshore false alarms, thus leading to higher detection performance. The experiments based on two measured SAR ship detection datasets demonstrate the effectiveness of the proposed method.

Convolutional neural networks (CNNs) have been widely used for synthetic aperture radar (SAR) target recognition. However, most existing SAR target recognition methods based on CNN only use amplitude information, and the extracted feature maps are difficult to be interpreted. For SAR target, the unique electromagnetic scattering characteristics are related to its physical characteristics, such as the shape, structure and material of the target. It is of great significance to design a physically interpretable SAR target recognition network combined with electromagnetic scattering characteristics. Considering that attributed scattering centers (ASCs) can effectively characterize the electromagnetic scattering characteristics of target and component information is robust to the local variations of target, we propose to divide SAR target in the light of the geometric scattering types of ASCs and construct a multi-scale CNN to comprehensively utilizes the global information and component information of the target. Due to the utilization of ASCs, the extracted feature maps of our method are more interpretable than those extracted by traditional CNNs. Experimental results on the MSTAR dataset prove the superior performance of our method.

In this letter, a target detector based on gradient harmonized mechanism (GHM) and attention mechanism is proposed to realize synthetic aperture radar (SAR) target detection in complex scenes. Considering the imbalance of positive and negative examples in SAR target detection, we use RefineDet as our backbone network. RefineDet can mitigate this imbalance problem by introducing the idea of two-step classification and regression into the one-stage detector. However, RefineDet only selects a part of examples for training and does not make full use of the information of all examples. Therefore, we apply GHM to the classification loss function of RefineDet, so that the network can make full use of all examples and increase the weights of hard examples adaptively in the loss function to reduce the false alarms and the missing alarms. In addition, to achieve a better detection performance in SAR images with complex scenes, a multiscale feature attention module (MFAM) is embedded into the network. By applying channel and spatial attention mechanisms to the multiscale feature maps, the MFAM can highlight the significant information and suppress the interference caused by clutter. The extensive experimental results based on the measured SAR dataset verify the effectiveness of the proposed method.

The existing deep learning studies on synthetic aperture radar (SAR) automatic target recognition (ATR) mainly focus on the utilization of the amplitude of SAR image via convolutional neural network (CNN), while the electromagnetic scattering information is rarely considered. Given that scattering centers (SCs) can characterize the target’s electromagnetic scattering characteristics and describe the target’s physical structure information, the SC feature should be helpful for SAR ATR. Therefore, we propose a novel SAR ATR method that combines electromagnetic scattering information and graph convolutional network (GCN) effectively and directly. Specifically, we model each extracted SC as a node to convert the SCs into graph data. The constructed graph is learned via GCN to describe the target’s physical structure information, where the features of different GCN layers are fused to avoid the over-smoothing of GCN. Label smoothing is combined with GCN for the first time to alleviate the overfitting caused by the limited training data. To the best of our knowledge, this study is the first to introduce the GCN for effectively utilizing the SCs, proving that the structural characteristics of the SCs of SAR targets are remarkably beneficial for recognition. Extensive experimental results on the measured moving and stationary target acquisition and recognition (MSTAR) dataset show that our method can obtain superior recognition performance compared with the existing methods.

This letter presents a generalized likelihood ratio test (GLRT)-based adaptive detector specially for the sub-Gaussian symmetric alpha-stable (SGS$\alpha \text{S}$) sea clutter background. Since the probability density function (PDF) of SGS$\alpha \text{S}$distribution cannot be expressed as a closed-form expression in terms of the elementary function, the research on target detectors in alpha-stable clutter background is very limited. In this letter, firstly, the Fox’s H-function is adopted to formulate the PDF of the SGS$\alpha \text{S}$distribution so that the PDF of SGS$\alpha \text{S}$distribution can be expressed as a closed form in terms of H-function. Then, the GLRT-based detector in alpha-stable clutter is designed based on the two-step GLRT criterion and the closed-form expression of the test statistics in terms of the H-function is given. Experimental results based on the simulated data and the measured data verify the effectiveness of the proposed detector.

In this article, a deep probability model, called the discriminative mixture variational autoencoder (DMVAE), is developed for the feature extraction in semisupervised learning. The DMVAE consists of three parts: 1) the encoding; 2) decoding; and 3) classification modules. In the encoding module, the encoder projects the observation to the latent space, and then the latent representation is fed to the decoding part, which depicts the generative process from the hidden variable to data. In particular, the decoding module in our DMVAE partitions the observed dataset into some clusters via multiple decoders whose number is automatically determined via the Dirichlet process (DP) and learns a probability distribution for each cluster. Compared to the standard variational autoencoder (VAE) describing all data with a single probability function, the DMVAE has the capacity to give a more accurate description for observations, thus improving the characterization ability of the extracted features, especially for the data with complex distribution. Moreover, to obtain a discriminative latent space, the class labels of labeled data are introduced to restrict the feature learning via a softmax classifier, with which the minimum entropy of the predicted labels for the features from unlabeled data can also be guaranteed. Finally, the joint optimization of the marginal likelihood, label, and entropy constraints makes the DMVAE have higher classification confidence for unlabeled data while accurately classifying the labeled data, ultimately leading to better performance. Experiments on several benchmark datasets and the measured radar echo dataset show the advantages of our DMVAE-based semisupervised classification over other related methods.

This article proposes a multiscale convolutional neural network (CNN) based on component analysis (CA-MCNN) for synthetic aperture radar (SAR) automatic target recognition (ATR). The component information of a target is robust to the local variations of the target, which is not made the best of by traditional CNN-based methods. For learning the component information, we use the attributed scattering centers (ASCs) extracted from the target echoes as the components of the target for SAR ATR, which divides the SAR target according to the geometric scattering types of ASCs and can not only make the division results more robust but also accurately characterize the electromagnetic scattering characteristics of the target. Since the global information provided by the whole image is also important for SAR ATR, CA-MCNN combines the global information with component information to learn a more efficient and robust target feature representation. In addition, considering that the feature maps of the shallower layer in CNN focus on local and fine-grained information while the feature maps in the deeper layer focus on global and coarse-grained information, we fuse the multiscale feature maps obtained from different layers to enhance the feature description ability. Extensive experiments conducted on the moving and stationary target acquisition and recognition (MSTAR) data set prove the superior performance of CA-MCNN.

In recent years, Synthetic Aperture Radar (SAR) ship detection methods based on convolutional neural networks have attracted wide attention in remote sensing fields. However, these methods require a large number of labeled SAR images to train the network, where labeling for SAR images is more expensive and time-consuming than that for optical images. To address the problem of lacking labeled SAR images, in this paper, we proposed an unsupervised domain adaptation framework based on progressive transfer for SAR ship detection by transferring knowledge from the optical domain to the SAR domain. Due to the prominent difference between the optical and SAR images, our approach progressively transfers knowledge from three levels: pixel level, feature level and prediction level. At the pixel level, considering the difference in imaging mechanism, we propose a special data augmentation method for ship targets and build the generator with skip-connection based on generative adversarial networks (GANs) to generate transition domain, which can reduce the appearance discrepancy between the optical and SAR images. At the feature level, the detector is trained to learn the domain-invariant features with adversarial alignment. At the prediction level, we further use the predicted pseudo-labels obtained by the feature-aligned detector to learn more discriminative features of the SAR images directly and propose the robust self-training (RST) method to reduce the influence of noisy pseudo-labels on detector training. Specially, RST is formulated as a loss minimization problem for object detection. The experimental results based on the domain adaptation from optical dataset to SAR dataset demonstrate that our approach achieves superior SAR ship detection performance with unlabeled SAR images.

The attitude and size of satellite targets are essential information for their activity analysis. This article proposes a novel approach to estimate the absolute attitude and size of satellite targets in the 3-D stable coordinates based on inverse synthetic aperture radar (ISAR) image interpretation. In an ISAR image of a satellite, the satellite’s body is chosen as an individual structure segmented from each ISAR image by employing pix2pix generative adversarial network (Pix2pixGAN). By exploring the shape feature of the satellite body with principal component analysis (PCA), the satellite attitude and size are estimated jointly through solving an optimization based on the gradient iteration method. The optimization is established by bridging range-Doppler (RD) images and the target feature parameters (attitude and size) with the accommodation of target trajectory information and the ISAR geometric projection model. In the experiments, the simulation data are generated from real satellite orbital parameters and the computer-aided-design (CAD) models of two satellite targets: TianGong (TG) and KeyHole (KH). Compared with the factorization-based reconstruction method, the proposed method can estimate the attitude and size of the satellite simultaneously and has a higher size estimation accuracy.

This article presents a novel approach for the joint estimation of satellite attitude and size based on inverse synthetic aperture radar (ISAR) image interpretation and parametric optimization. The satellite’s solar panel, which is segmented from the ISAR image by employing pix2pix generative adversarial network (Pix2pixGAN), is chosen for investigation in this article due to its unique rectangular structure. We innovatively use the principal component analysis (PCA) to explore the satellite solar panel’s structural features in an ISAR imagery. The projection matrix is then established to link the extracted features and the satellite’s absolute attitude and size. Parametric optimization is established based on the relationship between the extracted features and the satellite’s absolute attitude and size. A Broyden–Fletcher–Goldfarb–Shanno (BFGS)-based fast iterative search algorithm is employed to search the satellite’s absolute attitude and size simultaneously through an iterative approach. The simulation data are generated from actual satellite orbital parameters and computer-aided-design (CAD) models of the Aqua satellite in the experiments. Simulation experiments verify the effectiveness of the proposed method.

Three-dimensional (3-D) imaging plays a vital role in the recognition of spinning space debris. However, the image may be blurred due to the range migration caused by fast rotation of space debris. Moreover, the image quality, which depends on the estimation accuracy of Doppler frequency and chirp rate of scattering centers, is influenced by cross-terms and sidelobes. In this paper, we propose a novel 3-D imaging method based on smoothed Lv distribution (SLVD). Firstly, the selection criterion for best imaging time based on the time-frequency moment is proposed to guarantee that the echo is approximated as a linear frequency modulation signal. Then, we operate the Khatri-Rao product on the centroid frequency and chirp rate (CFCR) representation and the range-Doppler (RD) image to obtain the 3-D image. To decrease the influence of range migration, we process a short time window during the RD imaging procedure. For cross-term and sidelobe suppression, the SLVD is proposed to obtain the CFCR representation by expressing the Lv distribution (LVD) in a convolution form and introducing a centroid frequency window. Experimental results verify the effectiveness of the proposed imaging method and the good performance of the proposed SLVD for cross-term and sidelobe suppression.

In this study, a multi-scale convolutional factor analysis model with max-margin constraint (MMCFA) is developed for synthetic aperture radar (SAR) automatic target recognition. Compared with the traditional factor analysis (FA) model, the CFA model can maintain the spatial correlation among the image pixels in two-dimensional space and capture the structural information from images via convolution kernels. Moreover, multi-scale convolution kernels are adopted to capture richer features at different scales of SAR. Since the unsupervised model may not offer discriminative factors for the SAR target recognition task, it is expected to introduce the supervised information to the multi-scale CFA model when supervised information is available. Thus, a latent variable support vector machine (LVSVM) is linked to the factors learned from multi-scale CFA model, yielding max-margin discrimination, learned with the multi-scale CFA model jointly in a united framework. Experimental results on MSTAR dataset show that the proposed model has excellent recognition performance.

The Single Shot multi-box Detector (SSD) has been successfully applied in synthetic aperture radar (SAR) target detection. Besides, the saliency information in the saliency map has the ability that strengthening the target of interest and suppressing the clutter. It will help to improve the capability of scene understanding. According to the above, a novel SAR target detection network based on saliency-combined SSD is proposed. The proposed method includes two backbone sub-networks, one taking the SAR images as input for extracting the features, and the saliency map obtained from traditional saliency method is used as the input of the other sub-network to obtain the refined saliency information. Through the fusion module is used in multiple scales to integrate the saliency information and the network feature. Finally, we can get the detection results by the convolutional predictors on the multi-scale integrated feature maps. In addition, we apply the dense connection structure in the two sub-networks to utilize context information. The experimental results based on the miniSAR real data show that the proposed method can achieve a good detection performance.

To design a highly automatic and practical discrimination method for high-resolution synthetic aperture radar (SAR) images in complex scenes, a novel target discrimination framework based on weakly supervised learning (WSL) of the mid-level features is proposed in this article. First, we extract the dense SAR scale-invariant feature transform (SAR-SIFT) features of the candidate regions obtained from the detected SAR images. Then, the dense SAR-SIFT descriptors are transformed into richer mid-level features by coding and pooling. Finally, the mid-level features are input into a WSL-based target discrimination method, where the training set is initially selected by the unsupervised latent Dirichlet allocation (LDA) and iteratively updated by the linear support vector machine (SVM) discriminator. In the proposed method, only the image-level annotations (weak labels), which indicate whether the images containing the targets of interest or not, are required. By introducing WSL, the manual annotations of target regions from SAR images can be avoided, which is generally expensive in complex scenes and may tend to be less accurate and unreliable for the occluded or camouflaged targets. The comprehensive and specific experiments on the measured SAR data have demonstrated the effectiveness of the proposed method in benchmarking with the supervised learning-based linear SVM and linear support vector data description (SVDD) discriminators.

The single shot multibox detector (SSD), a proposal-free method based on convolutional neural network (CNN), has recently been proposed for target detection and has found applications in synthetic aperture radar (SAR) images. Moreover, the saliency information reflected in the saliency map can highlight the target of interest while suppressing clutter, which is beneficial for better scene understanding. Therefore, in this article, we propose a saliency-guided SSD (S-SSD) for target detection in SAR images, in which we effectively integrate the saliency into the SSD network not only to suggest where to focus on but also to improve the representation capability in complex scenes. The proposed S-SSD contains two separated convolutional backbone subnetwork architectures, one with the original SAR image as input to extract features, and the other with the corresponding saliency map obtained from the modified Itti's method as input to acquire refined saliency information under supervision. In addition, the dense connection structure, instead of the plain structure used in original SSD, is applied in the two convolutional backbone architectures to utilize multiscale information with fewer parameters. Then, for integrating saliency information to guide the network to emphasize informative regions, multilevel fusion modules are utilized to merge the two streams into a unified framework, thereby making the whole network end-to-end jointly trained. Finally, the convolutional predictors are used to predict targets. The experimental results on the miniSAR real data demonstrate that the proposed S-SSD can achieve better detection performance than state-of-the-art methods.

Synthetic aperture radar (SAR) target discrimination is usually performed in a supervised manner. However, supervised methods may suffer from lack of labeled training chips, whose acquirement is costly, time-consuming, and sometimes impossible. Moreover, traditional discrimination features only provide rough and partial description about chip and perform badly in SAR images with complex scenes. In order to solve these problems, we propose a novel semisupervised target discrimination method for SAR image by combining the feature learning with classifier learning into a uniform Bayesian framework based on a modified latent Dirichlet allocation (LDA) model, semisupervised infinite latent Dirichlet allocation (SSILDA). In our method, the semisupervised idea is used to deal with the difficulty of obtaining lots of labeled chips. A new variable is introduced into the LDA model for semisupervised learning, making it possible to obtain the semantic information of chips, while implementing the target discrimination in a semantic level. Moreover, the Dirichlet process (DP) is introduced into the LDA model to automatically determine the number of topics, and the parameters are inferred via Gibbs sampling. We have analyzed the performance of the proposed method comprehensively and specifically by using some measured data and carried out comparisons with the existing methods. The results validate the effectiveness of the proposed method for SAR target discrimination.

The traditional target detection methods for high-resolution synthetic aperture radar (SAR) images rely on the high intensity contrast between the targets and clutter, which are usually effective on the homogeneous background; however, they may lose effectiveness in the heterogeneous background. Compared with the clutter, the targets of interest usually have specific size and shape characteristics in high-resolution SAR images. Based on what mentioned above, we propose a target detection method based on the hierarchical saliency (HS) for high-resolution SAR images. The proposed method constructs the Bayesian saliency map to obtain the complete structures of both the targets of interest and strong clutter with superpixels as units firstly. Then, the morphological saliency map is constructed to retain the targets of interest while suppressing the strong clutter via the introduction of prior size information. The experimental results on the MiniSAR images show the effectiveness of the proposed target detection method.

In synthetic aperture radar (SAR) target discrimination, supervised methods, which need lots of labeled samples, are widely used. However, obtaining sufficient labeled samples is usually costly and laborious. In this paper, we propose a semi-supervised SAR target discrimination method based on co-training, which exploits both labeled and unlabeled samples. Firstly, we extract Lincoln features from the training samples and separate them into two subsets according to their physical meaning. Then, the co-training algorithm is utilized to iteratively train two support vector machine (SVM) classifiers on the extracted two feature subsets. Finally, the trained classifiers are exploited to classify the test data. The proposed method is evaluated with the miniSAR real dataset. Experimental results demonstrates that the proposed method is not only superior to the supervised methods with little labeled samples, but also the self-learning which only uses one feature subset.

In this letter, the single shot multibox detector (SSD), which is a real-time object detection method based on convolutional neural network, is applied to realize target detection for synthetic aperture radar (SAR) images. Since there are no sufficient labeled images for training in SAR target detection, we apply two strategies, data augmentation and transfer learning. For data augmentation, the first approaches to use some image processing methods, i.e., manual-extracting subimages, adding noise, filtering, and flipping, on the original training images to generate some new training images; the second approach is to employ the existing SAR target recognition data set, MSTAR data set, to assist in accomplishing the target detection task. For transfer learning, we first apply subaperture decomposition technique on original SAR images to acquire three-channel subaperture SAR images, and then transfer the three-channel VGGNet model pretrained on the ImageNet data set to the three-channel subaperture SAR images, in order to initialize corresponding parameters of the convolutional layers in the base network in our SSD. The feature extraction network, consisting of the base network and the auxiliary structure, is used to learn multiscale feature maps, and then convolutional predictors are used to acquire the final detection results. The experimental results on the miniSAR real image data set demonstrate that the proposed method can obtain better detection performance than other detection methods.

When a target is involved in maneuvering motion, the nonuniform 3-D rotation motion will cause a continuous change of image projection plane (IPP), which would induce 2-D spatial-variant phase errors. In this case, the inverse synthetic aperture (ISAR) image would be seriously blurred when using the traditional compensation methods. On the other hand, strong noise has been always challenging the conventional methods in motion parameters estimation and phase error compensation. In this paper, we propose a noise-robust compensation method to compensate the 2-D spatial-variant phase errors of the maneuvering target via using tracking information and parametric minimum entropy optimization. First, the maneuvering signal model is developed based on a 2-D spatial-variant model and a 3-D rotation motion model. Based on the developed signal model, a parametric entropy minimum optimization is established to estimate the rotation motion parameters. A gradient-based solver of this optimization is then adopted to iteratively find the global optimum. Meanwhile, in order to increase the robustness of this optimization under low SNR, an extended Kalman filter is adopted here for coarse motion estimation via using tracking information. By treating these estimated motion parameters as initial values, we can effectively prevent this optimization from trapping into a local optimum. Finally, the 2-D spatial-variant phase error can be iteratively compensated, and a well-focused ISAR image can be obtained. The proposed method has three main contributions: 1) it is applicable in the case of changing IPP; 2) it gives the exact expression of chip parameters; and 3) it can efficiently compensate the 2-D spatial-variant phase errors under low SNR. Experiments based on the simulated data and the real measured data prove the effectiveness and robustness of the proposed method.

Enhancing the contrast of target and clutter is a crucial issue in synthetic aperture radar (SAR) image target detection. In this paper, we define a novel subspace, called minimal clutter-to-signal ratio (MCSR) subspace, which can minimize the clutter-to-signal ratio (CSR) by projecting the feature vector to the subspace. Based on MCSR, a novel polarimetric contrast enhancement method is proposed. The MCSR subspace is learned based on the commonly used polarimetric feature vectors extracted from the labeled training SAR image pixels. The feature vectors extracted form candidate SAR image pixels are projected to the MCSR subspace. By calculating the square norm of each transformed feature vector, an enhanced image can be obtained. It is demonstrated that the existing optimization of polarimetric contrast enhancement (OPCE) is a special case of the proposed method to some extent. Experimental results show that our method outperforms the traditional OPCE method on the RadarSat-2 SAR data.

The conventional methods for target detection and discrimination in high-resolution synthetic aperture radar (SAR) images usually have low accuracy and slow speed, especially for large complex scenes. To overcome these drawbacks, in this paper, we propose a target detection and discrimination method based on visual attention model. In the detection stage, to pop out the targets and suppress the background clutter in the saliency map, we select the task-dependent scales from the Gaussian pyramid of the original SAR image. Moreover, we adopt the clustering algorithm to remerge several isolated focus of attention areas, which are obtained from the saliency map, into a complete target region. The candidate target SAR image chips are extracted with relative high accuracy and low time cost in this stage. Since there may be single target, multiple targets, or partial targets with complex clutter in each SAR image chip, it is hard to acquire accurate target-shaped blob via segmentation. Some classical discrimination features which are extracted based on target segmentation may lose effectiveness. In the discrimination stage of our method, to solve the above problem, based on the saliency and gist (SG) features for optical satellite images, we propose the modified SG (MSG) features for SAR target discrimination. The MSG features are complementary to each other and can provide a more complete description of the extracted SAR image chips without segmentation, which also reduces the computation burden. The experimental results on the synthetic images and miniSAR real SAR image data set demonstrate that the proposed target detection and discrimination method can detect and discriminate the targets from the complex background clutter with high accuracy and fast speed in high-resolution SAR images.

The classical target detection methods in synthetic aperture radar (SAR) images are mainly dependent on the intensity differences between the targets and clutter. Although they are effective in the simple scenes with high signal-to-clutter ratio (SCR), they may lose effectiveness in the complex scenes with low SCR. Generally, in high-resolution SAR images, the targets present not only high intensities but also specific size characteristics compared with the clutter. Based on this fact, in this paper, we propose a new target detection method for high-resolution SAR images via Bayesian-morphological saliency, which mainly contains two stages: Bayesian saliency map construction and morphological saliency map construction. The Bayesian saliency map can obtain the complete structures of the bright objects including the targets of interest and some bright clutter, via the superpixel segmentation and Bayesian framework. Furthermore, the morphological saliency map can highlight the targets of interest while suppressing both the natural and man-made clutter via the size prior information of the targets. The experimental results on the miniSAR real data set show that the proposed target detection method is effective.

In this paper, a novel noise-robust recognition method for high-resolution range profile (HRRP) data is proposed based on target scatterer pattern to enhance its recognition performance under the test condition of low SNR. The target dominant scatterers are first extracted based on the scattering center model of complex HRRP data via the orthogonal matching pursuit (OMP) algorithm to realize noise reduction. Then a scatterer matching recognition algorithm based on Hausdorff distance (HD) is developed with the magnitudes and locations of extracted dominant scatterers used as the feature patterns. Experimental results on the measured HRRP data demonstrate that the proposed method can improve the recognition performance under the relatively low SNR condition for both orthogonal and superresolution representations of scattering center model.

Target detection for synthetic aperture radar (SAR) images has great influence on the successive discrimination based on the target regions. However, as a pixel-based method, the traditional constant false alarm rate (CFAR) detection could not work well for the ship target detection problem of multiple ship targets with different sizes in a SAR image, which is referred to as the multiscale situation. Moreover, it needs to use the clustering method on the pixel-level detection results to obtain the accurate target regions, which may merge two or more different targets into a target region. In this letter, a modified CFAR based on object proposals is proposed. We use the object proposal generator to generate a small set of object proposals with different sizes, and then use the proposal-based CFAR detector, where the extracted object proposals are regarded as the guard windows instead of setting fixed guard window, to detect the true positive object proposals. By introducing the object proposals as the variable guard windows in the CFAR detector, the proposed algorithm could gain good detection performance in the multiscale situation, since the missed detection resulting from the big differences between the sizes of the fixed guard window and ship targets can be avoided. Meanwhile, the proposed method can directly obtain the accurate target regions. The effectiveness of the proposed algorithm is verified using the measured SAR data.

In this letter, we propose a novel automatic target recognition (ATR) method based on dictionary learning and joint dynamic sparse representation (DL-JDSR) for synthetic aperture radar (SAR) images. First, in the feature extraction step, we extract two kinds of features, i.e., the image domain amplitude feature and the scale-invariant feature transform (SIFT) feature, of which the image domain amplitude feature describes intensity information and the SIFT feature describes gradient information. These two features will be jointly utilized to combine the two kinds of information for SAR ATR. Second, we introduce the dictionary-learning method, the label-consistent K-singular value decomposition, into the training step to learn dictionaries for the two features rather than directly using all training samples as the fixed dictionaries in the traditional sparse representation method. The learned dictionaries have smaller sizes and are more distinctive among different classes, which can speed up our recognition and improve the accuracies. Third, the JDSR algorithm used in the test step employs a more flexible atom selection method, which enables the two features from an image data to share the similar but not exactly the same sparse mode. Experiments on the moving and stationary target acquisition and recognition data set show that the proposed method is an effective way to recognize SAR images.

This letter presents a new unsupervised distribution-free change detection method for synthetic aperture radar (SAR) images based on scale-invariant feature transform (SIFT) keypoints and region information. Since the SIFT can detect blob-like structures in an image and be insensitive to noise, we first extract noise-robust SIFT keypoints in the log-ratio image to reduce the detection range. Then, in order to obtain accurate changed regions, rather than directly obtaining the change-detection map from the difference image as in some traditional change detection methods, we make segmentation around the extracted keypoints in the two original multitemporal SAR images, where the edges of detection regions are much clearer than those in the difference image, and further compare the two segmentations to generate the change-detection map. This method utilizes the blob-like structure information offered by SIFT keypoints and the region information extracted via image segmentation. Experiments on real SAR images demonstrate the effectiveness of the proposed method.

A Bayesian classifier for sparsity-promoting feature selection is developed in this paper, where a set of nonlinear mappings for the original data is performed as a pre-processing step. The linear classification model with such mappings from the original input space to a nonlinear transformation space can not only construct the nonlinear classification boundary, but also realize the feature selection for the original data. A zero-mean Gaussian prior with Gamma precision and a finite approximation of Beta process prior are used to promote sparsity in the utilization of features and nonlinear mappings in our model, respectively. We derive the Variational Bayesian (VB) inference algorithm for the proposed linear classifier. Experimental results based on the synthetic data set, measured radar data set, high-dimensional gene expression data set, and several benchmark data sets demonstrate the aggressive and robust feature selection capability and comparable classification accuracy of our method comparing with some other existing classifiers.

In this letter, we introduce a novel noise-robust modification method for Gaussian-based models to enhance the performance of radar high-resolution range profile (HRRP) recognition under the test condition of low signal-to-noise ratio (SNR), and we develop an efficient scheme for its computation. This noise-robust modification method is implemented by revising the trained Gaussian-based model according to the estimated SNR of test HRRP. We apply the proposed method to adaptive Gaussian classifier and truncated stick-breaking hidden Markov model. Experimental results demonstrate that the proposed method can significantly improve the average recognition rate for noisy HRRP test samples while offering recognition performance comparable to that of original model for clean HRRP test samples. Moreover, even when the SNR of test HRRP samples is not precisely estimated, we can still obtain an acceptable result with the proposed method.

A novel method is proposed for moving wheeled vehicle and tracked vehicle classification using micro-Doppler features from returned radar signals within short dwell time. In this method, an adaptive analysis technique called Empirical Mode Decomposition (EMD) is utilized to decompose the motion components of moving vehicles, and a hierarchical classification structure using the decomposition results of returned signals is proposed to discriminate the two kinds of vehicles. The first stage of the structure elementarily identifies the tracked vehicle data by checking the existence of its unique feature and a further classification via our proposed features based on EMD is implemented in the second stage by using Support Vector Machine (SVM) classifier. Experimental results based on the simulated data and measured data are presented, including the performance analysis for low signal-to-noise ratio (SNR) case, generalization evaluation for different target circumstances and comparison with some related methods.

A non-parametric Bayesian model is proposed for processing multiple images. The analysis employs image features and, when present, the words associated with accompanying annotations. The model clusters the images into classes, and each image is segmented into a set of objects, also allowing the opportunity to assign a word to each object (localized labeling). Each object is assumed to be represented as a heterogeneous mix of components, with this realized via mixture models linking image features to object types. The number of image classes, number of object types, and the characteristics of the object-feature mixture models are inferred non-parametrically. To constitute spatially contiguous objects, a new logistic stick-breaking process is developed. Inference is performed efficiently via variational Bayesian analysis, with example results presented on two image databases.