Compressed sensing (CS) reconstruction of nonsparse scenes is one of the difficulties in synthetic aperture radar (SAR) imaging technology. Although the conventional CS method with sparse representation has proven applicable for nonsparse SAR reconstruction, its disadvantages are unsatisfactory imaging quality and high computational complexity under downsampling. In this letter, a novel deep learning approach for nonsparse SAR scene reconstruction is proposed based on sparse representation and the iterative shrinkage threshold algorithm (ISTA). Specifically, we first develop a sparse-representation-based imaging model associated with the $\ell _{1}$ sparse regularizer in nonlinear transform domains. Then, the advantages of the recurrent neural network (RNN) and convolutional neural network (CNN) are incorporated into an ISTA-inspired deep unfolded network (DUN) called SR-ISTA-Net, in which all the parameters are layer-varied rather than handcrafted. The experiments verify that the proposed SR-ISTA-Net can provide high-quality reconstruction results under nonsparse scenes while substantially reducing the imaging time.

Compressed sensing (CS)-based synthetic aperture radar (SAR) imaging methods are showing superior potential in imaging performance over classical matched filtering-based methods. However, the CS-based methods require much more computational cost to solve the iterative optimization composed of large-scale matrix operators. To hold the improvement of imaging performance and reduce the computational cost, in this article, we propose a novel SAR imaging method by a deep unfolded network (DUN) of an iterative shrinkage threshold algorithm (ISTA) with the approximated observation of a range-Doppler algorithm (RDA) operator. The proposed method takes the radar echoes as the input to learn the imaging procedure. First, the approximated observation is utilized in an SAR imaging model to reduce the size of the DUN. Moreover, we use ISTA as an example to introduce how to establish DUN with approximated observation, in which the detailed structure to handle the complex-valued radar echoes is also designed. Finally, the autoencoder is utilized to calculate the difference of the echoes rather than the imaging results, so that we can train the proposed network by unsupervised learning. The experiments of point targets, surface targets, and real scenes show that the proposed imaging method is superior in terms of imaging performance and computing efficiency.

Synthetic aperture radar (SAR) imaging has developed rapidly in recent years. Although the traditional sparse optimization imaging algorithm has achieved effective results, its shortcomings are slow imaging speed, large number of parameters, and high computational complexity. To solve the above problems, an end-to-end SAR deep learning imaging algorithm is proposed. Based on the existing SAR sparse imaging algorithm, the SAR imaging model is first rewritten to the SAR complex signal form based on the real-value model. Second, instead of arranging the two-dimensional echo data into a vector to continuously construct an observation matrix, the algorithm only derives the neural network imaging model based on the iteration soft threshold algorithm (ISTA) sparse algorithm in the two-dimensional data domain, and then reconstructs the observation scene through the superposition and expansion of the multi-layer network. Finally, through the experiment of simulation data and measured data of the three targets, it is verified that our algorithm is superior to the traditional sparse algorithm in terms of imaging quality, imaging time, and the number of parameters.

Traditional synthetic aperture radar (SAR) target detection methods use matched filtered (MF) SAR images as input, and the detection performance is restricted due to the high side lobes and speckle noise of these images. Sparse SAR imaging methods developed in recent years provide the advantages of reducing side lobes, noise and clutter. The imaging results obtained with these methods could help improve the SAR target detection performance. In this paper, to improve the target detection performance using sparse SAR images as input, we proposed a convolutional sparse feature enhancement method to meet the needs of Bayesian saliency detection. The proposed Bayesian saliency joint target detection method comprised the following three steps: first, to obtain sparse SAR images with continuous contours and fewer holes in the target area, we proposed a convolutional L1 sparse regularization method. Second, a regularization parameter optimization method was derived to quickly obtain optimal regularization parameters for saliency detection. Finally, target detection results were obtained through a superpixel-based Bayesian saliency joint detector. Extensive experiments verified that the proposed method could improve the SAR target detection accuracy in complex backgrounds.

Envelope alignment is one of the key steps for inverse synthetic aperture radar (ISAR) translational compensation. The traditional envelope alignment method cannot be accurately completed under a low signal-to-noise ratio (SNR), which will limit the accuracy of subsequent phase focusing. We propose a deep recurrent neural network (RNN) frame to address the problem. This is an end-to-end learning approach. Radar echo pulses are input to the network one by one according to time sequence. The inputs of each layer can be divided into two parts. The one is the current pulse, and the other one, named “state,” is the outputs of the previous layer except for the aligned pulse. Moreover, the outputs of each layer contain the “state” for the next layer and the aligned result of the input pulse. The above structure is a typical RNN, and the “states” transform the time-sequence information between different pulses. Compared with the traditional methods, the experiments verify that the proposed network can not only provide better alignment accuracy under low SNR but also require a shorter alignment time.

In Synthetic aperture radar (SAR) recognition tasks, due to its special imaging principle, SAR images acquired from different viewpoints contain target features that may carry a large amount of information. These target features from different viewpoints enrich the training data and also improve the target recognition performance. However, considering the relative difficulty of acquiring SAR images and their fine annotations, if recognition is forced by fusion of multiview features when raw data is scarce, feature redundancy will be formed, which in turn will lead to a decrease in recognition accuracy. To remedy the above problem, a multiview inter-class dissimilarity feature fusion (MIDFF) network is proposed in this study. The proposed network can simultaneously input images of two classes, one of which is a multiview input used to extract features from the same target at different views, and the other is input from other classes, which we call Input-heterogeneous, used to extract heterogeneous features. Firstly, a method for rapidly generating sufficient training data for MIDFF is proposed, which generates training data by repeatedly combining images from different views and classes to ensure that a large number of training inputs are available even when raw SAR images are scarce. Secondly, a method of calculating and enhancing of inter-class dissimilarity (ICD) features is proposed to increase the inter-class distance and improve the inter-class separability. Then, the ICD and multiview features are fused to increase the features learned by the network and reduce feature redundancy. Finally, a multiview heterogeneous weighted loss function is proposed, which combines the calculation of inter-class similarity and heterogeneous loss. Through the gradual convergence of the loss function, the inter-class similarity decreases as the loss decreases, which further improves the target recognition performance. Experimental results on MSTAR and Civilian Vehicle SAR datasets show that our proposed method performs better than the state-of-the-art methods within sample scarcity conditions.