MSTAR Dataset Research Articles

SAR Target Recognition Based on Modified Sparse Representation for Ground Safety

Recognition of synthetic aperture radar (SAR) targets is a hot topic in pattern recognition field. In the previous works, the sparse representation-based classification (SRC) is successfully used in SAR target recognition with high performance. The traditional SRC is performed on the global dictionary from the training classes. As a result, the representation capability of an individual class is not fully considered. This paper modifies the traditional SRC by performing the sparse representation over the local dictionaries formed by individual classes. In this way, the reconstruction error from one class can better reflect its representation capability as for describing the test sample. By comparing the reconstruction errors of different training classes, the target label of test sample can be classified finally. In the experiments, the MSTAR dataset is used to test the proposed method, which show the good results of the proposed method.

D-ATR for SAR Images Based on Deep Neural Networks

Automatic target recognition (ATR) can obtain important information for target surveillance from Synthetic Aperture Radar (SAR) images. Thus, a direct automatic target recognition (D-ATR) method, based on a deep neural network (DNN), is proposed in this paper. To recognize targets in large-scene SAR images, the traditional methods of SAR ATR are comprised of four major steps: detection, discrimination, feature extraction, and classification. However, the recognition performance is sensitive to each step, as the processing result from each step will affect the following step. Meanwhile, these processes are independent, which means that there is still room for processing speed improvement. The proposed D-ATR method can integrate these steps as a whole system and directly recognize targets in large-scene SAR images, by encapsulating all of the computation in a single deep convolutional neural network (DCNN). Before the DCNN, a fast sliding method is proposed to partition the large image into sub-images, to avoid information loss when resizing the input images, and to avoid the target being divided into several parts. After the DCNN, non-maximum suppression between sub-images (NMSS) is performed on the results of the sub-images, to obtain an accurate result of the large-scene SAR image. Experiments on the MSTAR dataset and large-scene SAR images (with resolution 1478 × 1784) show that the proposed method can obtain a high accuracy and fast processing speed, and out-performs other methods, such as CFAR+SVM, Region-based CNN, and YOLOv2.

Fusing Deep Learning and Sparse Coding for SAR ATR

We propose a multimodal and multidiscipline data fusion strategy appropriate for automatic target recognition (ATR) on synthetic aperture radar imagery. Our architecture fuses a proposed clustered version of the AlexNet convolutional neural network with sparse coding theory that is extended to facilitate an adaptive elastic net optimization concept. Evaluation on the MSTAR dataset yields the highest ATR performance reported yet, which is 99.33% and 99.86% for the three- and ten-class problems, respectively.

Synthetic Aperture Radar Target Recognition Based on Multidimensional Sparse Model

This paper introduces a novel classification strategy based on multi-dimensional sparse model for target recognition in Synthetic Aperture Radar (SAR) image. This method constructs dictionaries for each dimensions of SAR image instead of a single dictionary in traditional sparse model. In dictionary learning stage, this method adopts a dictionary-by-dictionary updating strategy. Under the premise of keeping the other dictionaries unchanged, each dictionary is updated using a traditional dictionary learning method. The sparsity coefficients of test samples in various categories of dictionaries are first calculated separately, and the test samples are classified using the sparse representation minimum error criterion. The experimental results on the MSTAR dataset show that compared with the traditional sparse models, this method can improve the target recognition rate while reducing the volume of dictionary. Therefore, in the field of SAR target identification, multidimensional sparse models have better recognition capabilities than traditional sparse models.

SAR Target Detection Based on SSD With Data Augmentation and Transfer Learning

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.

A Gradually Distilled CNN for SAR Target Recognition

Convolutional neural networks (CNNs) have been widely used in synthetic aperture radar (SAR) target recognition. Traditional CNNs suffer from expensive computation and high memory consumption, impeding their deployment in real-time recognition systems of SAR sensors, as these systems have low memory resources and low speed of calculation. In this paper, a micro CNN (MCNN) for real-time SAR recognition system is proposed. The proposed MCNN has only two layers, and it is compressed from a deep convolutional neural network (DCNN) with 18 layers by a novel knowledge distillation algorithm called gradual distillation. MCNN is a ternary network, and all its weights are either -1 or 1 or 0. Following a student-teacher paradigm, the DCNN is the teacher network and MCNN is its student network. The gradual distillation makes MCNN a better learning route than traditional knowledge distillation. The experiments on the MSTAR dataset show that the proposed MCNN can obtain a high recognition rate which is almost the same as the DCNN. However, compared with the DCNN, the memory footprint of the proposed MCNN is compressed 177 times, and the calculated amount is 12.8 times less, which means that the proposed MCNN can obtain better performance with the smaller network.

High-Performance SAR Automatic Target Recognition Under Limited Data Condition Based on a Deep Feature Fusion Network

Deep learning algorithms have been widely applied to the recognition of remote-sensing images due to their excellent performance on various recognition problems with sufficient data. However, limited data on synthetic aperture radar (SAR) images degrade the performance of neural networks for SAR automatic target recognition (ATR). To address this problem, this paper presents a new deep feature fusion framework by combining the Gabor features and information of raw SAR images and fusing the feature vectors extracted from different layers in our proposed neural network. Gabor features improve the richness of SAR image features. The number of free parameters of neural networks is largely reduced by utilizing large-scale convolutional kernel factorization and global average pooling. Moreover, the fusion of feature vectors from different layers helps improve the recognition performance of neural networks. Experimental results on the MSTAR dataset demonstrate the effectiveness of our proposed method. The proposed neural network can achieve an average accuracy of 99% on the classification of ten-class targets and even achieve a high recognition accuracy on limited data and noisy data.

Ground Target Classification in Noisy SAR Images Using Convolutional Neural Networks

Speckle noise is an inherent but annoying property in the synthetic aperture radar (SAR) imaging. In this paper we investigate the influence of speckle on the classical convolutional neural network (CNN) for SAR target classification. Then a dual stage coupled CNN architecture, named despeckling and classification coupled CNNs (DCC-CNNs), is proposed to distinguish multiple categories of ground targets in SAR images with strong and varying speckle. It first applies the despeckling sub-network for noise reduction. After that, residual speckle features as well as target information would be learned by the classification sub-network in order to solve the noise robustness problem of CNN. Besides, a new quantitative measure is developed for the quality assessment of SAR target images. It takes into account structural properties of the speckled SAR image of the target of interest and consistency with visual perception. Finally, a series of comparative experiments and discussions are carried out to validate the proposed assessment criterion and DCC-CNNs. Using synthetic SAR images based on the public MSTAR datasets, results show that the overall classification accuracy for ten ground target classes could be higher than 82% at a variety of speckle noise levels.

Multi-Stream Convolutional Neural Network for SAR Automatic Target Recognition

Despite the fact that automatic target recognition (ATR) in Synthetic aperture radar (SAR) images has been extensively researched due to its practical use in both military and civil applications, it remains an unsolved problem. The major challenges of ATR in SAR stem from severe data scarcity and great variation of SAR images. Recent work started to adopt convolutional neural networks (CNNs), which, however, remain unable to handle the aforementioned challenges due to their high dependency on large quantities of data. In this paper, we propose a novel deep convolutional learning architecture, called Multi-Stream CNN (MS-CNN), for ATR in SAR by leveraging SAR images from multiple views. Specifically, we deploy a multi-input architecture that fuses information from multiple views of the same target in different aspects; therefore, the elaborated multi-view design of MS-CNN enables it to make full use of limited SAR image data to improve recognition performance. We design a Fourier feature fusion framework derived from kernel approximation based on random Fourier features which allows us to unravel the highly nonlinear relationship between images and classes. More importantly, MS-CNN is qualified with the desired characteristic of easy and quick manoeuvrability in real SAR ATR scenarios, because it only needs to acquire real-time GPS information from airborne SAR to calculate aspect differences used for constructing testing samples. The effectiveness and generalization ability of MS-CNN have been demonstrated by extensive experiments under both the Standard Operating Condition (SOC) and Extended Operating Condition (EOC) on the MSTAR dataset. Experimental results have shown that our proposed MS-CNN can achieve high recognition rates and outperform other state-of-the-art ATR methods.

Pseudo-Zernike Moments Based Sparse Representations for SAR Image Classification

We propose radar image classification via pseudo-Zernike moments based sparse representations. We exploit invariance properties of pseudo-Zernike moments to augment redundancy in the sparsity representative dictionary by introducing auxiliary atoms. We employ complex radar signatures. We prove the validity of our proposed methods on the publicly available MSTAR dataset.

SAR Targets Classification Based on Deep Memory Convolution Neural Networks and Transfer Parameters

Deep learning has obtained state-of-the-art results in a variety of computer vision tasks and has also been used to solve SAR image classification problems. Deep learning algorithms typically require a large amount of training data to achieve high accuracy. In contrast, the size of SAR image datasets is often comparatively limited. Therefore, this paper proposes a novel method, deep memory convolution neural networks (M-Net), to alleviate the problem of overfitting caused by insufficient SAR image samples. Based on the convolutional neural networks (CNN), M-Net adds an information recorder to remember and store samples’ spatial features, and then it uses spatial similarity information of the recorded features to predict unknown sample labels. M-Net's use of this information recorder may cause difficulties for convergence if conventional CNN training methods were directly used to train M-Net. To overcome this problem, we propose a transfer parameter technique to train M-Net in two steps. The first step is to train a CNN, which has the same structure as the part of CNN incorporated in M-Net, to obtain initial training parameters. The second step applies the initialized parameters to M-Net and then trains the entire M-Net. This two-step training approach helps us to overcome the nonconvergence issue, and also reduces training time. We evaluate M-Net using the public benchmark MSTAR dataset, and achieve higher accuracy than several other well-known SAR image classification algorithms.

SAR Unlabeled Target Recognition Based on Updating CNN With Assistant Decision[-.2pc

Compared with optical images, the number of synthetic aperture radar (SAR) images is limited and growing slowly over time, and most of time the additional SAR images are unlabeled, which restricts the development of target recognition on SAR images. In this letter, the update learning method for SAR unlabeled target recognition based on the convolutional neural network (CNN) with assistant decision is proposed. The seed image set, which contains a small amount of existing labeled SAR images, is used for pretraining a CNN model. The unlabeled samples are fed to both CNN and the assistant classifier, and then, the matrices of the unlabeled samples based on recognition probability will be obtained, respectively. The unlabeled samples with high confidence by decision will be fed to CNN again to fine-tune the CNN model. The proposed method can make full use of the unlabeled samples to update the recognition model and to improve the recognition performance. Experiments on MSTAR data set show that, with the increase of unlabeled samples, the proposed method can improve the recognition performance and outperform the methods of CNN self-update learning and support vector machine update learning with the principal component analysis and the nonnegative matrix factorization. The experiments’ results demonstrate that the proposed method can effectively solve the problems caused by the lack of training samples.

Convolutional Neural Network-based SAR Image Classification with Noisy Labels

SAR image classification is an important task in SAR image interpretation. Supervised learning methods, such as the Convolutional Neural Network (CNN), demand samples that are accurately labeled. However, this presents a major challenge in SAR image labeling. Due to their unique imaging mechanism, SAR images are seriously affected by speckle, geometric distortion, and incomplete structural information. Thus, SAR images have a strong non-intuitive property, which causes difficulties in SAR image labeling, and which results in the weakened learning and generalization performance of many classifiers (including CNN). In this paper, we propose a Probability Transition CNN (PTCNN) for patch-level SAR image classification with noisy labels. Based on the classical CNN, PTCNN builds a bridge between noise-free labels and their noisy versions via a noisy-label transition layer. As such, we derive a new CNN model trained with a noisily labeled training dataset that can potentially revise noisy labels and improve learning capacity with noisily labeled data. We use a 16-class land cover dataset and the MSTAR dataset to demonstrate the effectiveness of our model. Our experimental results show the PTCNN model to be robust with respect to label noise and demonstrate its promising classification performance compared with the classical CNN model. Therefore, the proposed PTCNN model could lower the standards required regarding the quality of image labels and have a variety of practical applications.

SAR Automatic Target Recognition Based on Euclidean Distance Restricted Autoencoder

Deep learning algorithms have been introduced into target recognition of synthetic aperture radar (SAR) images for extracting deep features because of its accuracy on various recognition problems with sufficient training samples. However, applying deep structures in recognizing SAR images may suffer lack of training samples. Therefore, a deep learning method is proposed in this study based on a multilayer autoencoder (AE) combined with a supervised constraint. We bind the original AE algorithm with a restriction based on Euclidean distance to use the limited training images well. Moreover, a dropout step is added to our algorithm, which is designed to prevent overfitting caused by supervised learning. Experimental results on the MSTAR dataset demonstrate the effectiveness of the proposed method on real SAR images.

Aspect invariant features for radar target recognition

In this study, the authors introduce a novel method of target feature extraction that in one form attempts to reduce variability inherent in target signatures and in a second form attempts to exploit the variability. It is well known that radar target signatures can be subject to significant fluctuations as a function of viewing geometry and that this behaviour severely degrades the performance of automatic target recognition algorithms. Here, they propose one variability reduction technique which groups strong scatterers in target signatures. A second method exploits variability by using the difference between two reduced resolution images to extract target angular stability information. In both cases, a grid cell structure enables extraction of these relatively stable features from synthetic aperture radar images with low computational complexity. Results, using measured target data taken from the MSTAR dataset, demonstrate that the proposed method of selecting feature vectors significantly improve the overall classification performance. A correct classification rate of 96% is reached in testing.

Automatic Target Recognition of Military Vehicles With Krawtchouk Moments

The challenge of automatic target recognition of military targets within a synthetic aperture radar scene is addressed in this paper. The proposed approach exploits the discrete-defined Krawtchouk moments, which are able to represent a detected extended target with few features, allowing its characterization. The proposed algorithm provides robust performance for target recognition, identification, and characterization, with high reliability in the presence of noise and reduced sensitivity to discretization errors. The effectiveness of the proposed approach is demonstrated using the MSTAR dataset.

Object recognition using real and modelled SAR images

In this study, we examine our recognition approach using modelled SAR images as a test set. The supposed approach is compared with the technology of training using real data from the MSTAR dataset. We conducted the recognition experiments using three different objects. The SAR images were modelled with approximate parameters, such as those used in SAR constructions of real targets. As a recognition method, one proposed in our previous articles, the support subspaces method, is used. The experiment results show the ability to achieve high recognition quality with construction of the support subspaces (training) using modelled SAR images.

SAR Imagery Feature Extraction Using 2DPCA-Based Two-Dimensional Neighborhood Virtual Points Discriminant Embedding

Synthetic aperture radar (SAR) is an important microwave sensor that is capable of high-resolution imaging. Extracting valuable features from the SAR target imagery is one of crucial issues in SAR automatic target recognition (ATR). In this paper, we propose a new feature extraction method named 2-D principal-component-analysis-based 2-D neighborhood virtual points discriminant embedding (2DPCA-based 2DNVPDE) for SAR ATR. The SAR imagery is projected into the feature space by 2DPCA and 2DNVPDE in this approach. 2DPCA is able to preserve the global spatial structure of the original imagery, while 2DNVPDE establishes the spatial relationships of the neighborhoods to find the classification information from the neighborhoods of the samples. Hence, our method can extract powerful recognition information and represent the original image in low dimensions. Based on the MSTAR dataset, the experimental results show that the proposed method is able to achieve a higher recognition rate with a lower feature dimension over some existing SAR imagery feature extraction methods. Besides, it indicates that our method has a significant advantage in recognition performance and a lower sensitivity in statistical standpoint.

Holism-based features for target classification in focused and complex-valued synthetic aperture radar imagery

Reductionism and holism are two worldviews underlying the fields of linear and nonlinear signal processing, respectively. Conventional radar resolution theory is motivated by the former view, and it is violated by nonlinear phase modulation induced by dispersive scattering typically associated with extended targets. Motivated by the latter view, this paper offers a new insight into the process of feature extraction for target-recognition applications in single-channel imagery output from synthetic aperture radar processors. Two novel frameworks for holism-based feature extraction are presented. The first framework is based solely on the often-ignored phase chip. The second framework uses the complex-valued 2-D synthetic aperture radar chip after it is transformed into a 1-D vector. Representative features are introduced under each framework. Further, for comparison purposes, baseline features from the power-detected chip are also considered. Three feature sets are extracted from the real-world MSTAR data set and used separately and combinatorially to design multiple instances of an eight-class support vector machine classifier. A classification accuracy of 93.42% is achieved for the holism-based features. This is in comparison to 73.63% for the baseline features. Using Fisher scoring to measure the information contained in each feature, top-ranked features from the first and second holism-based frameworks, respectively, are found to be 7 and 160 times those of the baseline features. Because the nonlinear phenomenon is resolution dependent, our proposed approach is expected to achieve even greater accuracy for synthetic aperture radar sensors with higher resolution.

SAR 영상을 이용한 템플릿 매칭 기반 자동식별 알고리즘 구현 및 성능시험

In this paper, we have developed a target recognition algorithm based on a template matching technique using Synthetic Aperture Radar (SAR) images. For efficient computations, Radon transform-based azimuth estimation algorithm was used with the template matching. MSTAR data set was divided into two groups according to the depression angles, which were a train set and a test set. Template data were generated by rotating and cropping chips which were from MSTAR train set using the azimuth estimation algorithm. Then the template matching process between test data and template data was performed under various conditions. Performance variation according to contrast enhancement preprocessing which is scarce in open literature was also presented. The analysis results show that the target recognition algorithm could be useful for the automatic target recognition using SAR images.