Deep Convolution Network with Sparse Prior for Sparse ISAR Image Enhancement

In view of the serious transverse fringe interference caused by the sparsity of data echo in inverse synthetic aperture radar (ISAR) sparse imaging, a deep learning based ISAR sparse imaging enhancement technology is proposed. Firstly, the regularization sparsity constraint is introduced into the network loss function by using the prior information of ISAR image sparsity. The purpose is to improve the performance of the network to suppress false targets and reduce the image sidelobe. At the same time, the sparse feature extraction module composed of dilated convolution and standard convolution is helpful to obtain advanced semantic information, extract more context semantic features, and enhance the feature expression ability of the network, which helps to further improve the network's reconstruction quality of sparse ISAR images. Our proposed method has the advantages of high quality of reconstructed images compared with other traditional methods. Experimental results show the effectiveness of the method.