SAR Target Classification Based on Sample Spectral Regularization

Synthetic aperture radar (SAR) is a microwave imaging equipment based on the principle of synthetic aperture, which has all kinds of characteristics such as all-time, all-weather, high resolution and wide breadth. It also has high research value and applied foreground in the area of military and civilian. In particular, worldwide, a great deal of researches on SAR target classification and identification based Deep Learning are ongoing, and the obtained results are highly effective. However, it is well known that Deep Learning requires a large amount of data, and it is costly and inaccessible to acquire SAR samples through field experiment, so image simulation research for expanding SAR dataset is essential. In this paper, we concentrated on generating highly realistic SAR simulated images for several equipment models using Generative Adversarial Network (GAN) without construction of terrain scene model and RCS material mapping. Then we tested the SAR simulated images on a specialized SAR classification model pretrained on MSTAR dataset. The results showed that simulated targets could be identified and classified accurately, demonstrating the high similarity of SAR simulated images with real samples. Our work could provide a greater variety of available SAR images for target classification and identification study.

Deep learning networks are widely being applied to remote sensing image recognition and have achieved promising results. In this paper, we researched the influence of background with different scattering characteristics for synthetic aperture radar (SAR) target recognition based on convolutional neural network (CNN). Firstly, a two-parameter CFAR image segmentation method based on Weibull distribution was used to extracted SAR target and its shadow. And then, SAR datasets with road, farmland and grassland background environment is synthesized to analyze the CNN classifier. Experiments results show that the method by mixing training sets with different background together can improve the recognization rate when the backgrounds are complex.

This paper presents an enhanced ensemble classification framework for Synthetic Aperture Radar (SAR) Automatic Target Recognition (ATR) under diverse operational conditions, including Standard Operating Conditions (SOC) and Extended Operating Conditions (EOC). The proposed method integrates the strengths of Residual Neural Networks (ResNet) replacing Convolutional Neural Networks (CNN), Support Vector Machines (SVM), and template matching, leveraging majority voting to combine their complementary capabilities. The ensemble framework achieves improved robustness and classification accuracy across varied scenarios. The methodology employs ResNet, a deep learning architecture known for its superior feature extraction and classification capabilities, replacing AlexNet to address limitations in generalization and consistency. ResNet demonstrated better performance with average accuracies of 92.67% under SOC and 88.9% under EOC, showing consistent results across all six target classes, as compared to the CNN-based ensemble approach with average accuracies of 90.30% under SOC and 87.22% under EOC. The SVM is employed for its robustness in handling overfitting and classifying features extracted from 16 region properties. Template matching is included for its resilience in challenging conditions where deep learning techniques may underperform. Experimental validation using the MSTAR dataset, a standard benchmark for SAR ATR, highlights the effectiveness of this ensemble approach. The results confirm significant improvements in classification accuracy and robustness over individual classifiers, demonstrating the practical applicability of the ensemble approach to real-world SAR ATR challenges. This research advances SAR ATR by addressing critical challenges, including noise, occlusion, and variations in viewing angles while achieving high classification performance under diverse conditions. The integration of ResNet further enhances the framework’s adaptability and reliability.

A target detection method based on improved single shot multibox detector (SSD) is proposed to solve insufficient training samples for synthetic aperture radar (SAR) target detection. Firstly, a residual network is introduced in the feature extraction structure for extracting deep features of the target; then, the aspect ratio of default boxes is redesigned to match different sizes of target detection; finally, the existing dataset is expanded with small samples based on saliency map. The experimental results based on the simulated SAR data and the real SAR data prove that the proposed method has better performance in SAR target detection.

The current Synthetic Aperture Radar (SAR) target detection methods based on Convolutional Neural Network (CNN) rely on a large amount of slice-level labeled train samples. However, it takes a lot of labor and material resources to label the SAR images at slice-level. Compared to label samples at slice-level, it is easier to label them at image-level. The image-level label indicates whether the image contains the target of interest or not. In this paper, a semi-supervised SAR image target detection method based on CNN is proposed by using a small number of slice-level labeled samples and a large number of image-level labeled samples. The target detection network of this method consists of region proposal network and detection network. Firstly, the target detection network is trained using the slice-level labeled samples. After training convergence, the output slices constitute the candidate region set. Then, the image-level labeled clutter samples are input into the network and then the negative slices of the output are added to the candidate region set. Next, the image-level labeled target samples are input into the network as well. After selecting the positive and negative slices in the output of the network, they are added to the candidate region set. Finally, the detection network is trained using the updated candidate region set. The processes of updating candidate region set and training detection network alternate until convergence. The experimental results based on the measured data demonstrate that the performance of the proposed method is similar to the fully supervised training method using a much larger set of slice-level samples.

In the field of synthetic aperture radar (SAR) target recognition, leveraging computer-generated labeled simulated SAR data to aid in the recognition of unlabeled real SAR data has garnered significant attention. However, the domain shift between the simulated and real domain contravenes the assumption of independent and identically distributed in deep learning. Moreover, the categories of real SAR data may not align precisely with those of the simulated SAR data, and the prior knowledge of the label shift typically remains unavailable. To align the common category samples and identify target-private unknown category samples in the presence of both domain shift and potential label shift, we introduce a two-stage universal domain adaptation (UniDA) method for simulation-assisted SAR target recognition based on adversarial uncertainty and neighbor relation (AUNR). The proposed AUNR introduces Dirichlet distribution-based evidential deep learning, which not only performs classification tasks but also characterizes decision uncertainty, enabling the model to express the concept of “Unknow”. In the first stage, alongside supervised learning, we leverage data augmentation-based contrastive learning to achieve robust feature representation. In the second stage, adversarial learning based on model and data uncertainty is employed to separate common and unknown class samples, and contrastive learning based on neighbor relation aligns samples in two domains according to their geometric relationships. To the best of our knowledge, this is the first study on UniDA for SAR target recognition. Experimental results demonstrate that our method outperforms previous state of-the-art methods in various label shift scenarios, which indicates the effectiveness of the proposed method.

Detailed information on spatial distribution of wetland classes is crucial for monitoring this important productive ecosystem using advanced remote sensing tools and data. Although the potential of full- and dual-polarimetric (FP and DP) Synthetic Aperture Radar (SAR) data for wetland classification has been well examined, the capability of compact polarimetric (CP) SAR data has not yet been thoroughly investigated. This is of great significance, since the upcoming RADARSAT Constellation Mission (RCM), which will soon be the main source of SAR observations in Canada, will have CP mode as one of its main SAR configurations. This also highlights the necessity to fully exploit such important Earth Observation (EO) data by examining the similarities and dissimilarities between FP and CP SAR data for wetland mapping. Accordingly, this study examines and compares the discrimination capability of extracted features from FP and simulated CP SAR data between pairs of wetland classes. In particular, 13 FP and 22 simulated CP SAR features are extracted from RADARSAT-2 data to determine their discrimination capabilities both qualitatively and quantitatively in three wetland sites, located in Newfoundland and Labrador, Canada. Seven of 13 FP and 15 of 22 CP SAR features are found to be the most discriminant, as they indicate an excellent separability for at least one pair of wetland classes. The overall accuracies of 87.89%, 80.67%, and 84.07% are achieved using the CP SAR data for the three wetland sites (Avalon, Deer Lake, and Gros Morne, respectively) in this study. Although these accuracies are lower than those of FP SAR data, they confirm the potential of CP SAR data for wetland mapping as accuracies exceed 80% in all three sites. The CP SAR data collected by RCM will significantly contribute to the efforts ongoing of conservation strategies for wetlands and monitoring changes, especially on large scales, as they have both wider swath coverage and improved temporal resolution compared to those of RADARSAT-2.

Synthetic Aperture Radar (SAR) images are prone to be contaminated by noise, which makes it very difficult to perform target recognition in SAR images. Inspired by great success of very deep convolutional neural networks (CNNs), this paper proposes a robust feature extraction method for SAR image target classification by adaptively fusing effective features from different CNN layers. First, YOLOv4 network is fine-tuned to detect the targets from the respective MF SAR target images. Second, a very deep CNN is trained from scratch on the moving and stationary target acquisition and recognition (MSTAR) database by using small filters throughout the whole net to reduce the speckle noise. Besides, using small-size convolution filters decreases the number of parameters in each layer and, therefore, reduces computation cost as the CNN goes deeper. The resulting CNN model is capable of extracting very deep features from the target images without performing any noise filtering or pre-processing techniques. Third, our approach proposes to use the multi-canonical correlation analysis (MCCA) to adaptively learn CNN features from different layers such that the resulting representations are highly linearly correlated and therefore can achieve better classification accuracy even if a simple linear support vector machine is used. Experimental results on the MSTAR dataset demonstrate that the proposed method outperforms the state-of-the-art methods.

In order to improve synthetic aperture radar (SAR) target recognition performance, this paper proposes a novel method using multi-level deep features. The multi-level deep features are learned by the convolutional neural network (CNN), which are capable of describing the target characteristics from different aspects. In order to make full use of the discrimination contained in the multi-level deep features, the joint sparse representation (JSR) is used as the basic classifier, which performs the multi-task learning to jointly classify the multi-level deep features. It could not only represent each feature properly but also consider the correlations between different levels of features. Based on the solutions from JSR, the target label is classified as the training class with the minimum reconstruction error. By fully exploiting the discriminative information contained in the multi-level deep features, the proposed method could effectively enhance SAR target recognition performance. The moving and stationary target acquisition and recognition (MSTAR) dataset is employed in the experiments. The results show that the proposed method could achieve a significantly high recognition rate of 99.38% for classifying 10 classes of targets under the standard operating condition (SOC), which is higher than those from some reference methods drawn from current literatures. Under different types of extended operating conditions (EOCs), the overall performance of the proposed method keeps superior over the reference methods. In addition, the outlier rejection capability of the proposed method is also better than the compared methods. All these experimental results validate the high effectiveness of the proposed method.

Despite recent research on the potential of dual- (DP) and full-polarimetry (FP) Synthetic Aperture Radar (SAR) data for crop mapping, the capability of compact polarimetry (CP) SAR data has not yet been thoroughly investigated. This is of particular concern, given the availability of such data from RADARSAT Constellation Mission (RCM) shortly. Previous studies have illustrated potential for accurate crop mapping using DP and FP SAR features, yet what contribution each feature makes to the model accuracy is not well investigated. Accordingly, this study examined the potential of the early- to mid-season (i.e., May to July) RADARSAT-2 SAR images for crop mapping in an agricultural region in Manitoba, Canada. Various classification scenarios were defined based on the extracted features from FP SAR data, as well as simulated DP and CP SAR data at two different noise floors. Both overall and individual class accuracies were compared for multi-temporal, multi-polarization SAR data using the pixel- and object-based random forest (RF) classification schemes. The late July C-band SAR observation was the most useful data for crop mapping, but the accuracy of single-date image classification was insufficient. Polarimetric decomposition features extracted from CP and FP SAR data produced relatively equal or slightly better classification accuracies compared to the SAR backscattering intensity features. The RF variable importance analysis revealed features that were sensitive to depolarization due to the volume scattering are the most important FP and CP SAR data. Synergistic use of all features resulted in a marginal improvement in overall classification accuracies, given that several extracted features were highly correlated. A reduction of highly correlated features based on integrating the Spearman correlation coefficient and the RF variable importance analyses boosted the accuracy of crop classification. In particular, overall accuracies of 88.23%, 82.12%, and 77.35% were achieved using the optimized features of FP, CP, and DP SAR data, respectively, using the object-based RF algorithm.

Because of the special imaging mechanism of Synthetic Aperture Radar (SAR) and the existence of the speckle noise, SAR target configuration recognition has been a hard task in SAR target recognition. We propose a method of SAR target configuration recognition by using the Locality Preserving Projections (LPP), which is a subspace analytical method based on manifold learning. The proposed method extracts features from SAR images in the manifold space which is more suitable for the real SAR images than the Euclidean space. The feature extraction method by LPP not only preserves the global topology structure, but also captures the local information of the target with different configurations. Experimental results on MSTAR datasets suggest that the proposed method can provide a higher recognition rate in SAR target configuration recognition.

The science plan for the Alaska SAR Facility (ASF) focuses on earth surface characteristics that are of interest within the overall concept of global change and that show significant regional, seasonal and interannual variations resulting in changes in the strength of their radar returns. The polar oceans, with the continuous motion and deformation of the pack ice and the changes in the surface state of the surrounding open seas, offer excellent opportunities for such research. Because such studies require both frequent and detailed analysis of Synthetic Aperture Radar (SAR) data, a Geophysical Processor System (GPS) has been developed to speed the extraction of useful geophysical information from SAR data sets. The system will initially produce three main types of products: (a) sets of ice motion vectors obtained by automated computer tracking of identifiable ice floes on sequential images, (b) the areal extent and location of several different ice types and open water and (c) a characterization of the wave state in ice-free regions as well as within the ice in the marginal ice zone at locations where significant wave penetration occurs. Details of these analysis procedures are described. Initially the GPS is planned to process 10 image pairs/day for ice motion, 20 images/day for ice type variations and 1 image/day for wave information, with a total estimated processing time of 13 hours. A variety of projects plan to utilize the SAR data stream in studies of ice, lead and polynya dynamics and thermodynamics. A common feature of these research programs will be attempts to provide, via the coupling of the SAR data with ice property and ice dynamics models, improved estimates of the heat and mass fluxes into both the atmosphere and the ocean as affected by the characteristics of the ice cover.Key words: SAR, radar, sea ice, image analysis, remote sensing

In the field of earth sciences and remote exploration, the classification and identification of surface materials on earth have been a significant research area that poses considerable challenges in recent times. Although deep learning technology has achieved certain results in remote sensing image classification, it still has certain challenges for multi-modality remote sensing data classification. In this paper, we propose a fusion network based on transformer and gist convolutional neural network (CNN), namely TGF-Net. To minimize the duplication of information in multimodal data, the TGF-Net network incorporates a feature reconstruction module (FRM) that employs matrix factorization and self-attention mechanism for decomposing and evaluating the similarity of multimodal features. This enables the extraction of distinct as well as common features. Meanwhile, the transformer-based spectral feature extraction module (TSFEM) was designed by combining the different characteristics of remote sensing images and considering the problem of orderliness of the sequence between hyperspectral image (HSI) channels. In order to address the issue of representing the relative positions of spatial targets in synthetic aperture radar (SAR) images, we proposed a spatial feature extraction module called gist-based spatial feature extraction module (GSFEM). To assess the efficacy and superiority of the proposed TGF-Net, we performed experiments on two datasets comprising HSI and SAR data.

Synthetic-aperture radar (SAR) image target detection is widely used in military, civilian and other fields. However, existing detection methods have low accuracy due to the limitations presented by the strong scattering of SAR image targets, unclear edge contour information, multiple scales, strong sparseness, background interference, and other characteristics. In response, for SAR target detection tasks, this paper combines the global contextual information perception of transformers and the local feature representation capabilities of convolutional neural networks (CNNs) to innovatively propose a visual transformer framework based on contextual joint-representation learning, referred to as CRTransSar. First, this paper introduces the latest Swin Transformer as the basic architecture. Next, it introduces the CNN’s local information capture and presents the design of a backbone, called CRbackbone, based on contextual joint representation learning, to extract richer contextual feature information while strengthening SAR target feature attributes. Furthermore, the design of a new cross-resolution attention-enhancement neck, called CAENeck, is presented to enhance the characterizability of multiscale SAR targets. The mAP of our method on the SSDD dataset attains 97.0% accuracy, reaching state-of-the-art levels. In addition, based on the HISEA-1 commercial SAR satellite, which has been launched into orbit and in whose development our research group participated, we released a larger-scale SAR multiclass target detection dataset, called SMCDD, which verifies the effectiveness of our method.

Recently, Convolutional Neural Network (CNN) has achieved some success in synthetic aperture radar (SAR) image classification. This outstanding performance mainly depends on a large number of training samples, but achieved satisfactory classification results with limited training samples remains a challenge. To address this problem, we propose a dual-scale Siamese densely connected network with Markov Random Fields (DS-SDCNet-MRF) for single-polarization SAR image classification. First, the Siamese densely connected network (SDCNet) is proposed to fully extract discriminative features under limited samples. Then, the proposed DS-SDCNet is constructed with two SDCNets of different scales to produce complementary classification results. Among them, the large-scale SDCNet has a better classification result in the homogeneous region, while the small-scale SDCNet tends to provide good detail preservation. Finally, an improved MRF model which combined the category probability information is proposed to further improve the classification performance. Experimental results on simulated and real single-polarization SAR data demonstrate that the proposed method achieves more encouraging classification performance than the current state-of-the-art classification methods with limited samples.