Synthetic aperture radar automatic target recognition (SAR-ATR) has made great progress in recent years. Most of the established recognition methods are supervised, which have strong dependence on image labels. However, obtaining the labels of radar images is expensive and time-consuming. In this paper, we present a semi-supervised learning method that is based on the standard deep convolutional generative adversarial networks (DCGANs). We double the discriminator that is used in DCGANs and utilize the two discriminators for joint training. In this process, we introduce a noisy data learning theory to reduce the negative impact of the incorrectly labeled samples on the performance of the networks. We replace the last layer of the classic discriminators with the standard softmax function to output a vector of class probabilities so that we can recognize multiple objects. We subsequently modify the loss function in order to adapt to the revised network structure. In our model, the two discriminators share the same generator, and we take the average value of them when computing the loss function of the generator, which can improve the training stability of DCGANs to some extent. We also utilize images of higher quality from the generated images for training in order to improve the performance of the networks. Our method has achieved state-of-the-art results on the Moving and Stationary Target Acquisition and Recognition (MSTAR) dataset, and we have proved that using the generated images to train the networks can improve the recognition accuracy with a small number of labeled samples.

The deep convolution neural network (CNN), which has prominent advantages in feature learning, can learn and extract features from data automatically. Existing polarimetric synthetic aperture radar (PolSAR) image classification methods based on the CNN only consider the polarization information of the image, instead of incorporating the image’s spatial information. In this paper, a novel method based on a dual-branch deep convolution neural network (Dual-CNN) is proposed to realize the classification of PolSAR images. The proposed method is built on two deep CNNs: one is used to extract the polarization features from the 6-channel real matrix (6Ch) which is derived from the complex coherency matrix. The other is utilized to extract the spatial features of a Pauli RGB (Red Green Blue) image. These extracted features are first combined into a fully connected layer sharing the polarization and spatial property. Then, the Softmax classifier is employed to classify these features. The experiments are conducted on the Airborne Synthetic Aperture Radar (AIRSAR) data of Flevoland and the results show that the classification accuracy on 14 types of land cover is up to 98.56%. Such results are promising in comparison with other state-of-the-art methods.

The recent emergence of high-resolution Synthetic Aperture Radar (SAR) images leads to massive amounts of data. In order to segment these big remotely sensed data in an acceptable time frame, more and more segmentation algorithms based on deep learning attempt to take superpixels as processing units. However, the over-segmented images become non-Euclidean structure data that traditional deep Convolutional Neural Networks (CNN) cannot directly process. Here, we propose a novel Attention Graph Convolution Network (AGCN) to perform superpixel-wise segmentation in big SAR imagery data. AGCN consists of an attention mechanism layer and Graph Convolution Networks (GCN). GCN can operate on graph-structure data by generalizing convolutions to the graph domain and have been successfully applied in tasks such as node classification. The attention mechanism layer is introduced to guide the graph convolution layers to focus on the most relevant nodes in order to make decisions by specifying different coefficients to different nodes in a neighbourhood. The attention layer is located before the convolution layers, and noisy information from the neighbouring nodes has less negative influence on the attention coefficients. Quantified experiments on two airborne SAR image datasets prove that the proposed method outperforms the other state-of-the-art segmentation approaches. Its computation time is also far less than the current mainstream pixel-level semantic segmentation networks.

With the rapid development of deep learning technology, many synthetic aperture radar (SAR) target recognition algorithms based on convolutional neural networks have achieved exceptional performance on various datasets. However, conventional neural networks are repeatedly iterated on a fixed dataset until convergence, and once they learn new tasks, a large amount of previously learned knowledge is forgotten, leading to a significant decline in performance on old tasks. This article presents an incremental learning method based on strong separability features (SSF-IL) to address the model’s forgetting of previously learned knowledge. The SSF-IL employs both intraclass and interclass scatter to compute the feature separability loss, in order to enhance the linear separability of features during incremental learning. In the process of learning new classes, an intraclass clustering loss is proposed to replace the conventional knowledge distillation. This loss function constrains the old class features to cluster around the saved class centers, maintaining the separability among the old class features. Finally, a classifier bias correction method based on boundary features is designed to reinforce the classifier’s decision boundary and reduce classification errors. SAR target incremental recognition experiments are conducted on the MSTAR dataset, and the results are compared with several existing incremental learning algorithms to demonstrate the effectiveness of the proposed algorithm.