MSTAR Dataset Research Articles

A Multiscale Local–Global Feature Fusion Method for SAR Image Classification with Bayesian Hyperparameter Optimization Algorithm

In recent years, the advancement of deep learning technology has led to excellent performance in synthetic aperture radar (SAR) automatic target recognition (ATR) technology. However, due to the interference of speckle noise, the task of classifying SAR images remains challenging. To address this issue, a multi-scale local–global feature fusion network (MFN) integrating a convolution neural network (CNN) and a transformer network was proposed in this study. The proposed network comprises three branches: a CovNeXt-SimAM branch, a Swin Transformer branch, and a multi-scale feature fusion branch. The CovNeXt-SimAM branch extracts local texture detail features of the SAR images at different scales. By incorporating the SimAM attention mechanism to the CNN block, the feature extraction capability of the model was enhanced from the perspective of spatial and channel attention. Additionally, the Swin Transformer branch was employed to extract SAR image global semantic information at different scales. Finally, the multi-scale feature fusion branch was used to fuse local features and global semantic information. Moreover, to overcome the problem of poor accuracy and inefficiency of the model due to empirically determined model hyperparameters, the Bayesian hyperparameter optimization algorithm was used to determine the optimal model hyperparameters. The model proposed in this study achieved average recognition accuracies of 99.26% and 94.27% for SAR vehicle targets under standard operating conditions (SOCs) and extended operating conditions (EOCs), respectively, on the MSTAR dataset. Compared with the baseline model, the recognition accuracy has been improved by 12.74% and 25.26%, respectively. The results demonstrated that Bayes-MFN reduces the inter-class distance of the SAR images, resulting in more compact classification features and less interference from speckle noise. Compared with other mainstream models, the Bayes-MFN model exhibited the best classification performance.

Distillation Sparsity Training Algorithm for Accelerating Convolutional Neural Networks in Embedded Systems

The rapid development of neural networks has come at the cost of increased computational complexity. Neural networks are both computationally intensive and memory intensive; as such, the minimal energy and computing power of satellites pose a challenge for automatic target recognition (ATR). Knowledge distillation (KD) can distill knowledge from a cumbersome teacher network to a lightweight student network, transferring the essential information learned by the teacher network. Thus, the concept of KD can be used to improve the accuracy of student networks. Even when learning from a teacher network, there is still redundancy in the student network. Traditional networks fix the structure before training, such that training does not improve the situation. This paper proposes a distillation sparsity training (DST) algorithm based on KD and network pruning to address the above limitations. We first improve the accuracy of the student network through KD, and then through network pruning, allowing the student network to learn which connections are essential. DST allows the teacher network to teach the pruned student network directly. The proposed algorithm was tested on the CIFAR-100, MSTAR, and FUSAR-Ship data sets, with a 50% sparsity setting. First, a new loss function for the teacher-pruned student was proposed, and the pruned student network showed a performance close to that of the teacher network. Second, a new sparsity model (uniformity half-pruning UHP) was designed to solve the problem that unstructured pruning does not facilitate the implementation of general-purpose hardware acceleration and storage. Compared with traditional unstructured pruning, UHP can double the speed of neural networks.

Modern Synergetic Neural Network for Synthetic Aperture Radar Target Recognition.

Feature extraction is an important process for the automatic recognition of synthetic aperture radar targets, but the rising complexity of the recognition network means that the features are abstractly implied in the network parameters and the performances are difficult to attribute. We propose the modern synergetic neural network (MSNN), which transforms the feature extraction process into the prototype self-learning process by the deep fusion of an autoencoder (AE) and a synergetic neural network. We prove that nonlinear AEs (e.g., stacked and convolutional AE) with ReLU activation functions reach the global minimum when their weights can be divided into tuples of M-P inverses. Therefore, MSNN can use the AE training process as a novel and effective nonlinear prototypes self-learning module. In addition, MSNN improves learning efficiency and performance stability by making the codes spontaneously converge to one-hots with the dynamics of Synergetics instead of loss function manipulation. Experiments on the MSTAR dataset show that MSNN achieves state-of-the-art recognition accuracy. The feature visualization results show that the excellent performance of MSNN stems from the prototype learning to capture features that are not covered in the dataset. These representative prototypes ensure the accurate recognition of new samples.

A new model based on multi-aspect images and complex-valued neural network for synthetic aperture radar automatic target recognition

ABSTRACT During the last decades, Synthetic Aperture Radar Automatic Target Recognition (SAR ATR) has been vastly utilized in the military services and civil studies. Because of the sensibility of the SAR images of the imaging in the azimuth dimension, using the multi-aspect SAR image sequence is more practical than the single-aspect SAR image to achieve the superior classification accuracy in an actual SAR ATR task. At present time, multi-aspect SAR ATR models mostly utilize Recurrent Neural Networks (RNN) that depend on the sequence between samples and so suffer from the lack of information. In a practical work, a huge amount of training set is needed to train a deep learning network accurately, but it is so costly work to extract multi-aspect SAR images. So, in this article, a new model of multi-aspect SAR ATR is proposed by using a self-attention model. The proposed self-attention model is utilized to calculate the internal correlation between the original SAR images. At the same time, to develop the anti-noise capability of the proposed model and decrease the dependency on a huge amount of training data, a Convolutional Auto Encoder (CAE) is designed and utilized in the feature extraction section of the proposed model. On the other hand, unlike existing traditional methods, in this work we use both amplitude and phase information of SAR images to devolve the training process of the proposed model. It should be noted that all of the parameters of the proposed network are developed to a complex domain. In addition a complex backpropagation algorithm by using the gradient based model is used for training the network. In the end, experiments is obtained by using two MSTAR data set (the MSTAR-SOC and EOC). The experimental results prove that the proposed model not only can achieve a high recognitions rate on the case of sufficient training sample but also can obtain an acceptable rate in the case of small training samples in different complex situations. In addition, the simulation results demonstrate that by using the encoder of CAE in the proposed model, the whole configuration of the proposed model achieves the anti-noise capability that is a valuable benefit of any practical SAR ATR task.

SAR Target Recognition with Limited Training Samples in Open Set Conditions.

It is difficult to collect training samples for all types of synthetic aperture radar (SAR) targets. A realistic problem comes when unseen categories exist that are not included in training and benchmark data at the time of recognition, which is defined as open set recognition (OSR). Without the aid of side-information, generalized OSR methods used on ordinary optical images are usually not suitable for SAR images. In addition, OSR methods that require a large number of samples to participate in training are also not suitable for SAR images with the realistic situation of collection difficulty. In this regard, a task-oriented OSR method for SAR is proposed by distribution construction and relation measures to recognize targets of seen and unseen categories with limited training samples, and without any other simulation information. The method can judge category similarity to explain the unseen category. Distribution construction is realized by the graph convolutional network. The experimental results on the MSTAR dataset show that this method has a good recognition effect for the targets of both seen and unseen categories and excellent interpretation ability for unseen targets. Specifically, while recognition accuracy for seen targets remains above 95%, the recognition accuracy for unseen targets reaches 67% for the three-type classification problem, and 53% for the five-type classification problem.

Azimuth-Aware Discriminative Representation Learning for Semi-Supervised Few-Shot SAR Vehicle Recognition

Among the current methods of synthetic aperture radar (SAR) automatic target recognition (ATR), unlabeled measured data and labeled simulated data are widely used to elevate the performance of SAR ATR. In view of this, the setting of semi-supervised few-shot SAR vehicle recognition is proposed to use these two forms of data to cope with the problem that few labeled measured data are available, which is a pioneering work in this field. In allusion to the sensitivity of poses of SAR vehicles, especially in the situation of only a few labeled data, we design two azimuth-aware discriminative representation (AADR) losses that suppress intra-class variations of samples with huge azimuth-angle differences, while simultaneously enlarging inter-class differences of samples with the same azimuth angle in the feature-embedding space via cosine similarity. Unlabeled measured data from the MSTAR dataset are labeled with pseudo-labels from categories among the SARSIM dataset and SAMPLE dataset, and these two forms of data are taken into consideration in the proposed loss. The few labeled samples in experimental settings are randomly selected in the training set. The phase data and amplitude data of SAR targets are all taken into consideration in this article. The proposed method achieves 71.05%, 86.09%, and 66.63% under 4-way 1-shot in EOC1 (Extended Operating Condition), EOC2/C, and EOC2/V, respectively, which overcomes other few-shot learning (FSL) and semi-supervised few-shot learning (SSFSL) methods in classification accuracy.

Variable Length Sequential Iterable Convolutional Recurrent Network for UWB-IR Vehicle Target Recognition

A variable length sequential iterable convolutional recurrent network (VS-ICRN) is proposed in this paper, aiming at improving the vehicle target recognition ability for the Ultra-Wideband Impulse Radar (UWB-IR). Firstly, the array imaging technology is introduced into the UWB-IR, and thus a range-angle imaging method for the array UWB-IR is put forward, to simulate the array UWB-IR vehicles image under different observation conditions. Secondly, in order to make full use of both the deep features in the single image and the deep associated features between the sequence images, a VS-ICRN model is proposed, which includes three sub-modules: the image feature extraction based on the iterable convolution, the variable length sequential image associated feature extraction and the target classification, respectively. Finally, the experiment on the simulation dataset and the MSATAR is carried out to validate the effectiveness of the proposed method. The experimental results on the simulation dataset show that when SNR=-10dB, the proposed method is superior in the recognition rate to the GoogLeNet and AlexNet methods with 16% and 19%, respectively. Meanwhile, the proposed VS-ICRN method only needs 1.38% parameters quantity to achieve a comparable recognition rate as GoogLeNet on MSTAR dataset.

Discovering and Explaining the Noncausality of Deep Learning in SAR ATR

In recent years, deep learning has been widely used in SAR ATR and achieved excellent performance on the MSTAR dataset. However, due to constrained imaging conditions, MSTAR has data biases such as background correlation, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">i.e</i> ., background clutter properties have a spurious correlation with target classes. Deep learning can overfit clutter to reduce training errors. Therefore, the degree of overfitting for clutter reflects the non-causality of deep learning in SAR ATR. Existing methods only qualitatively analyze this phenomenon. In this paper, we quantify the contributions of different regions to target recognition based on the Shapley value. The Shapley value of clutter measures the degree of overfitting. Moreover, we explain how data bias and model bias contribute to non-causality. Concisely, data bias leads to comparable signal-to-clutter ratios and clutter textures in training and test sets. And various model structures have different degrees of overfitting for these biases. The experimental results of various models under standard operating conditions on the MSTAR dataset support our conclusions. Our code is available at https://github.com/waterdisappear/Data-Bias-in-MSTAR.

SM-CNN: Separability Measure-Based CNN for SAR Target Recognition

With the maturity of deep learning algorithm in Synthetic Aperture Radar (SAR) target recognition filed, Convolutional Neural Network (CNN) has become the most effective model. However, the interpretability and the separability of feature maps extracted from convolution layers have not been specially analyzed neither qualitatively nor quantitatively, which makes the traditional model work like a “black box”. To alleviate the problem, a novel model based on separability measure (SM) - CNN is proposed in this letter, which introduces the principle of maximal coding rate reduction to the backbone module. SM-CNN quantitatively analyzes the separability of the feature maps and takes the value as a vital part of the loss function to guide the training process of the model. The calculation process of the separability measure values can be strictly derived mathematically, so it is more interpretable, turning the black box into a “gray box”. Additionally, the proposed model can achieve comparable recognition performance of the backbone networks with reduced computational complexity. Comparative experiments based on MSTAR and OpenSARShip data sets verify the effectiveness and practicability of the method proposed in this letter.

PAN: Part Attention Network Integrating Electromagnetic Characteristics for Interpretable SAR Vehicle Target Recognition

Machine learning methods for synthetic aperture radar (SAR) image automatic target recognition (ATR) can be divided into two main types: traditional methods and deep learning methods. The deep learning methods can learn the high-dimensional features of the target directly, and usually obtain high target recognition accuracy. However, they lack full consideration of SAR targets’ inherent characteristics resulting in poor generalization and interpretation ability. Compared with the deep learning methods, traditional methods can get more interpretable and stable results with model-based features. In order to take full advantage of these two kinds of methods, we propose target part attention network based on the attributed scattering center (ASC) model to integrate the electromagnetic characteristics with the deep learning framework. Firstly, considering the importance of scattering structure for SAR ATR, we design a target part model based on ASC model. Then, a novel part attention module based on Scaled Dot-Product Attention mechanism is proposed, which directly associates the features of target parts with the classification results. Finally, we give the derivation method of the importance of each part, which is of great significance for practical application and the interpretation of SAR ATR. Experiments on the MSTAR data set demonstrate the effectiveness of the proposed part attention network. Compared with existing studies, it can achieve higher and more robust classification accuracy under different complex conditions. Furthermore, combined with the importance of parts, we constructed two effective interpretable analysis methods for deep learning network classification results.

Federated Target Recognition for Multiradar Sensor Data Security

Single sensor radar can no longer satisfy the increasingly complex electromagnetic environment. More attention is paid to radar sensor networks, which can obtain more information from different nodes and points of view. Moreover, better detection and tracking performance can be achieved through resource sharing and complementary advantages (joint learning). How to improve the utilization efficiency of multiple radar sensors with limited resources is an open problem, which can be transformed into joint learning in scenarios such as limited training data or imbalanced samples. This paper presents a distributed learning model to solve these problems. It has three phases. Self-reweighting loss is developed to dynamically rebalance the gradients of positive and negative samples for each category, after which the imbalance of samples can be alleviated. An image generation technique via target reimaging addresses the problem of limited samples. Self-reweighting loss and image generation are then unified in a federated learning framework. The classifier is adjusted using virtual representations to further improve learning efficiency. Comparative studies on the MSTAR dataset demonstrate the advantages of the proposed method.

Occluded Target Recognition in SAR Imagery With Scattering Excitation Learning and Channel Dropout

Deep neural networks are widely used in SAR image classification and recognition, achieving state-of-the-art performance. But it remains a challenging task to recognize occluded targets. In this letter, we propose a novel robust SAR recognition method against occlusion. Specifically, we design a scattering excitation learning module that encourages the network to learn more robust features responding to the scattering centers of targets. In addition, we adopt a random feature channel dropout technique which can further improve robustness to occlusion. Our method makes the network more robust against occlusion but without any occlusion-simulated data for training. Experimental results on MSTAR dataset shows that our proposed method achieves remarkably improved robustness even under severe occlusions. Code is made available at https://github.com/koervcor/SEL-CD.

Synthetic Aperture Radar Target Recognition Based on Multimodule Image Enhancement Network

Synthetic aperture radar (SAR) has been widely used in recent years, and SAR automatic target recognition (ATR) has become a research hotspot. Most of the existing SAR ATR methods focus on the network structure design and increasing data volume but omit image quality and real-time processing. We design a multimodule image enhancement network (MMIE-Net) to solve these problems, which include the target extraction module, the image processing module, and convolutional neural networks (CNNs). First, we use the target extraction module and the image processing module to enhance the quality of raw SAR images. Then we design a suitable network for SAR image recognition, which is simple, lightweight, and recognizable. The experiment was mainly carried out on the MSTAR dataset, which can be divided into two categories, Standard Operating Condition (SOC) and Extended Operating Condition (EOC). The identification accuracy, the parameter storage space, and the depth of the model are considered as the criterion. The experimental results show that, compared with other methods, the proposed method not only ensures the simple structure of the network model but also has better recognition accuracy. Additionally, our method is robust and stable to large depression angle variation as well.

SAR Target Recognition Based on Inception and Fully Convolutional Neural Network Combining Amplitude Domain Multiplicative Filtering Method

The research of Synthetic aperture radar (SAR) target recognition plays a significant role in military and civilian fields. However, for small sample SAR target recognition, there are some problems that need to be solved urgently, including low recognition accuracy, slow training convergence rate, and serious overfitting. Aiming at the above problems, we propose a recognition method based on Inception and Fully Convolutional Neural Network (IFCNN) combined with Amplitude Domain Multiplicative Filtering (ADMF) image processing. To improve the recognition accuracy and convergence rate, the ADMF method is utilized to construct the pretraining set, and the initial parameters of the network are optimized by pretraining. In addition, this paper builds the IFCNN model by introducing the Inception structure and the mixed progressive convolution layer into the FCNN. The full convolution structure of FCNN is effective to alleviate the problem of network overfitting. The Inception structure can enhance the sparsity of features and improve the network classification ability. Meanwhile, the mixed progressive convolution layers can accelerate training. Based on the MSTAR dataset, the experimental results show that the method proposed achieves an average precision of 88.95% and the training convergence rate is significantly improved in small sample scenarios.

SAR Target Recognition via Monogenic Signal and Gaussian Process Model

The monogenic signal and Gaussian process model are applied to synthetic aperture radar (SAR) target recognition. The monogenic signal is used to extract the features of the SAR image. The Gaussian process model is a statistical learning algorithm based on the Bayesian theory, which constructs a classification model by combining the kernel function and the probability judgement. Compared with the traditional classification model, the Gaussian process model can obtain higher classification efficiency and accuracy. During the implementation, the monogenic feature vector of the SAR image is used as the input, and the target label is used as the output to train the Gaussian process model. For the test sample to be classified, the target label is determined by calculating the posterior probability of each class using the Gaussian process model. In the experiments, the validations are carried out under typical conditions based on the MSTAR dataset. According to the experimental results, the proposed method maintains the highest performance under the standard operating condition, depression angle differences, and noise corruption, which verifies its effectiveness and robustness.

Semi-Supervised SAR ATR Framework with Transductive Auxiliary Segmentation

Convolutional neural networks (CNNs) have achieved high performance in synthetic aperture radar (SAR) automatic target recognition (ATR). However, the performance of CNNs depends heavily on a large amount of training data. The insufficiency of labeled training SAR images limits the recognition performance and even invalidates some ATR methods. Furthermore, under few labeled training data, many existing CNNs are even ineffective. To address these challenges, we propose a Semi-supervised SAR ATR Framework with transductive Auxiliary Segmentation (SFAS). The proposed framework focuses on exploiting the transductive generalization on available unlabeled samples with an auxiliary loss serving as a regularizer. Through auxiliary segmentation of unlabeled SAR samples and information residue loss (IRL) in training, the framework can employ the proposed training loop process and gradually exploit the information compilation of recognition and segmentation to construct a helpful inductive bias and achieve high performance. Experiments conducted on the MSTAR dataset have shown the effectiveness of our proposed SFAS for few-shot learning. The recognition performance of 94.18% can be achieved under 20 training samples in each class with simultaneous accurate segmentation results. Facing variances of EOCs, the recognition ratios are higher than 88.00% when 10 training samples each class.

An Empirical Study of Fully Black-Box and Universal Adversarial Attack for SAR Target Recognition

It has been demonstrated that deep neural network (DNN)-based synthetic aperture radar (SAR) automatic target recognition (ATR) techniques are extremely susceptible to adversarial intrusions, that is, malicious SAR images including deliberately generated perturbations that are imperceptible to the human eye but can deflect DNN inference. Attack algorithms in previous studies are based on direct access to a ATR model such as gradients or training data to generate adversarial examples for a target SAR image, which is against the non-cooperative nature of ATR applications. In this article, we establish a fully black-box universal attack (FBUA) framework to craft one single universal adversarial perturbation (UAP) against a wide range of DNN architectures as well as a large fraction of target images. It is of both high practical relevance for an attacker and a risk for ATR systems that the UAP can be designed by an FBUA in advance and without any access to the victim DNN. The proposed FBUA can be decomposed to three main phases: (1) SAR images simulation, (2) substitute model training, and (3) UAP generation. Comprehensive evaluations on the MSTAR and SARSIM datasets demonstrate the efficacy of the FBUA, i.e., can achieve an average fooling ratio of 64.6% on eight cutting-edge DNNs (when the magnitude of the UAP is set to 16/255). Furthermore, we empirically find that the black-box UAP mainly functions by activating spurious features which can effectively couple with clean features to force the ATR models to concentrate on several categories and exhibit a class-wise vulnerability. The proposed FBUA aligns with the non-cooperative nature and reveals the access-free adversarial vulnerability of DNN-based SAR ATR techniques, providing a foundation for future defense against black-box threats.

Multi-Aspect Convolutional-Transformer Network for SAR Automatic Target Recognition

In recent years, synthetic aperture radar (SAR) automatic target recognition (ATR) has been widely used in both military and civilian fields. Due to the sensitivity of SAR images to the observation azimuth, the multi-aspect SAR image sequence contains more information for recognition than a single-aspect one. Nowadays, multi-aspect SAR target recognition methods mainly use recurrent neural networks (RNN), which rely on the order between images and thus suffer from information loss. At the same time, the training of the deep learning model also requires a lot of training data, but multi-aspect SAR images are expensive to obtain. Therefore, this paper proposes a multi-aspect SAR recognition method based on self-attention, which is used to find the correlation between the semantic information of images. Simultaneously, in order to improve the anti-noise ability of the proposed method and reduce the dependence on a large amount of data, the convolutional autoencoder (CAE) used to pretrain the feature extraction part of the method is designed. The experimental results using the MSTAR dataset show that the proposed multi-aspect SAR target recognition method is superior in various working conditions, performs well with few samples and also has a strong ability of anti-noise.

Fast Target Detection Algorithm Based on CFAR and Target Variance Characteristics

Target detection is a complex process that is important as an important module in computer vision applications. In particular, in many occasions where the real-time requirements are extremely high, it is very important to achieve fast and accurate detection of targets. But at this stage, there are still many problems in the research on rapid target detection, such as inefficiency and high is the first phase of automatic target recognition (ATR). For the performance of SAR image target detection, this paper proposes a CFAR fast detection algorithm based on Rayleigh. CFAR detection is divided into two steps: horizontal and vertical CFAR detection. The efficiency of parameter estimation is improved by the coincidence of adjacent point reference windows and the distribution characteristics of images. The algorithm in this paper combines the target variance characteristics to reduce the false alarm rate. The experiment was performed on the MSTAR dataset. Fast target detection algorithm based on CFAR and target variance feature has the characteristics of high detection rate, low false alarm, and high speed, and its detection performance is good. The experimental results show that the recognition efficiency of the proposed algorithm is higher than that of the traditional algorithm on different target datasets, the time is shortened by 30%, and the accuracy rate is equal to that of the traditional algorithm.

Детектирование объектов военной техники с использованием сверточной нейронной сети на радиолокационных изображениях, сформированных в условиях шумовых помех

The objects detector on radar images description is given, which implements the local inhomogeneities detection using the CFAR algorithm with their subsequent two-alternative classification («object» or «background») by a deep convolutional neural networkю. Using the MSTAR dataset, it was shown that in the case of a homogeneous background, the detection results can be considered satisfactory if SNR on image is more than 5 dB, and completely disrupted if SNR less then minus 5 dB.