To improve the classification performance of the convolutional neural network (CNN) for polarimetric synthetic aperture radar (PolSAR) images with limited labeled samples, this letter proposes a memory CNN (MCNN) for semisupervised PolSAR image classification using both the labeled and unlabeled samples. Specifically, the MCNN introduces a memory module to realize an assimilation–accommodation interaction between the network and the module in the model training process. Compared with the traditional CNN-based methods, the advantage of the introduced interaction mechanism can exploit the memory information during the model training including both the learned feature representation and the model inference uncertainty. Under the framework of memory mechanism, the semisupervised learning can be implemented simply and effectively by introducing an unsupervised memory loss. We evaluate the proposed method on three benchmark PolSAR data sets. The experimental results show the advantages of the MCNN over the supervised, semisupervised, and unsupervised methods in the PolSAR image classification with limited labeled samples.

This paper presents a synthetic aperture radar (SAR) target recognition method based on meta knowledge transferring with relation comparison. Firstly, the meta knowledge is obtained based on a relation network by learning a distance metric from simulated data. Then, we retrain the network on the limited real SAR target samples to realize the meta knowledge transferring. Different from the conventional methods based on feature transferring, the meta knowledge learned from simulated data using relation network enables the model to adapt to the recognition task of real SAR data quickly, which is called “learning to learn”. Under the condition of limited training samples, the experimental results on MSTAR dataset verify the advantages of the proposed method compared with the conventional convolutional neural network (CNN)-based method.

Automatic target recognition (ATR) for synthetic aperture radar (SAR) has been a significant research topic for decades. Existing methods of ATR perform the task after image formation. However, the imaging process of target information may lose the target abstract feature information hidden in the original data. Motivated by this, we feed the SAR raw data to a deep convolutional neural network (CNN) that does not require image reconstruction. To evaluate the method, we trained two well-known CNN networks, VGG19 and ResNet18, on the simulated dataset and MSTAR dataset. The results show that both CNN network architectures can achieve good results for classification.

This paper proposes a SAR target recognition method based on enhanced discriminant feature learning. Specifically, a compactness constraint is added to the convolutional autoencoder (CAE) to minimizes the reconstruction loss and the intra-class sample distance, which results in a enhanced discriminant feature representation. Afterwards, we use the encoder of the pretrained CAE with compactness constraint to initialize a convolutional neural network (CNN) to construct an end-to-end recognition model. Experimental results on the moving and stationary target acquisition and recognition (MSTAR) dataset show that the proposed method achieves competitive performance with limited training samples.

Most of the existing methods for automatic target recognition (ATR) in synthetic aperture radar (SAR) images only focus on the seen classes in the testing stage, but ignore the possibility of the unseen classes. To address this problem, a generalized zero-shot SAR target recognition method based on conditional disentangled representation learning is proposed, which can be used for both seen and unseen classes. The recognition framework is based on the variational autoencoders (VAE) to disentangle the continuous and discrete latent variables. The unseen classes are recognized by continuous variables, while the discrete ones are used for the seen classes. In addition, the class labels of seen classes are incorporated as condition information to improve the inter-class discrimination in two disentangled variable spaces. The generalized zero-shot experiment results show that the proposed method achieves considerable recognition accuracy for both seen and unseen classes.

Learning discriminative features is difficult for deep learning-based target recognition in synthetic aperture radar (SAR) images with small training samples. To achieve a better feature learning, this study proposes a new deep network, a compact convolutional autoencoder (CCAE) for SAR target recognition. CCAE minimises the reconstruction loss and the distance between intra-class samples simultaneously by imposing compactness constraint on the encoder, which results in a more discriminative feature representation. Furthermore, the pretrained CCAE encoder can be used to initialise the corresponding parameters of a convolutional neural network to facilitate the training of the end-to-end model. Experimental results using the moving and stationary target acquisition and recognition dataset show that the proposed method outperforms the existing deep learning-based methods in the case of small training samples.

In the past decade, the compressive sensing (CS) based ISAR imaging methods have gained much interest due to the capability of obtaining high-quality images with under-sampled data. However, both the performance and the application of CS ISAR imaging methods are limited by sparse representations and iterative reconstruction algorithms where the former may be hard to find and the latter may lead to low imaging efficiency. The deep-network (DN) based image reconstruction methods appeared in recent years have been shown to be able to dramatically reduce the computational complexity and ensure the image reconstruction at the same time. The Deep-ADMM-Net (DAN) is a network constructed by iterative steps of the traditional optimization algorithm, Alternating Direction Method of Multipliers (ADMM). The mainly characteristic of DAN is that it is interpretable and efficient. We propose a DAN based ISAR imaging method. The well trained DAN can reconstruct high-quality ISAR image using much fewer measurements than CS based imaging methods. Experimental results show that our proposed imaging method is superior to the existing CS method in both image reconstruction quality and computational efficiency.

Because of the unknown target information and the complex time-varying environments, the uncertainty of target information is introduced into the radar echo model, and hence to reduce the influence of uncertainty of target information and save the antenna array elements. A novel grey chance constrained programming (GCCP) model is proposed for antenna aperture resource management of digital array radar. For solving the multi-object model, the grey simulation is integrated into fast and elitist nondominated sorting genetic algorithm (NSGA_II) to compose a hybrid optimization algorithm on the condition that the specified confidence levels are satisfied. Simulations are carried out to verify the effectiveness of the proposed GCCP model. Finally, the optimal solutions of antenna aperture allocation are obtained.

Land cover classification is an important part of the polarimetric synthetic aperture radar (PolSAR) image interpretation. The convolutional neural network (CNN) has been utilized to improve the classification accuracy recently. However, how to efficiently train the classification model with limited training samples while keeping the generalization performance is still a challenge. In this letter, we devise a subspace learning network (SSLNet) for PolSAR image classification, which can be trained more efficiently. First, a third-order polarimetric feature tensor is constructed using five-target decompositions to make full use of the prior knowledge. The tensor is then fed into a two-layer CNN in which the principal component analysis (PCA) is employed to learn the convolutional filters. Finally, the output features of the network are obtained by binary hashing and block-wise histograms, followed by the nearest neighbor (NN) classifier to complete the classification. Due to the simple learning strategy, the proposed SSLNet can be easily designed and efficiently trained. Experimental results on benchmark PolSAR data reveal that the SSLNet can achieve higher classification accuracy with limited training samples than the conventional CNN method.

To further improve the effect of infrared small target detection, a reweighted infrared patch-image model is proposed. First, the authors point out that the nuclear norm in the infrared patch-image model could easily leave some sparse background edges in the target patch-image, leading to an inaccurate background estimation. Then, to overcome this defect, the reweighted nuclear norm is adopted to constrain the background patch-image, which could preserve the background edges better. Considering that some non-target sparse points could not be suppressed by only using l(1) norm, the authors introduce the reweighted l(1) norm to further enhance the sparsity of target image. Finally, the proposed model is formulated as a reweighted robust principal component analysis problem and solved by the inexact augmented Lagrangian multiplier method. Extensive experiments show that the proposed model outperforms the other six competitive methods in suppressing background clutter and detecting target.

To further enhance the small targets and suppress the heavy clutters simultaneously, a robust non-negative infrared patch-image model via partial sum minimization of singular values is proposed. First, the intrinsic reason behind the undesirable performance of the state-of-the-art infrared patch-image (IPI) model when facing extremely complex backgrounds is analyzed. We point out that it lies in the mismatching of IPI model’s implicit assumption of a large number of observations with the reality of deficient observations of strong edges. To fix this problem, instead of the nuclear norm, we adopt the partial sum of singular values to constrain the low-rank background patch-image, which could provide a more accurate background estimation and almost eliminate all the salient residuals in the decomposed target image. In addition, considering the fact that the infrared small target is always brighter than its adjacent background, we propose an additional non-negative constraint to the sparse target patch-image, which could not only wipe off more undesirable components ulteriorly but also accelerate the convergence rate. Finally, an algorithm based on inexact augmented Lagrange multiplier method is developed to solve the proposed model. A large number of experiments are conducted demonstrating that the proposed model has a significant improvement over the other nine competitive methods in terms of both clutter suppressing performance and convergence rate.