Scanning Radar Scene Reconstruction With Deep Unfolded ISTA Neural Network

Abstract

Complex scene reconstruction is one of the most critical issues in scanning radar processing. The azimuth echo of the scanning radar can be equivalent to the convolution result of the scene scattering coefficient and the antenna pattern. Iter-ative shrinkage-thresholding algorithm (ISTA) has been proven effective in the target reconstruction of the scanning radar, but it often performs unsatisfactory reconstruction quality on complex scenes. This paper proposes a new learning-based approach, an improved ISTA-based deep unfolding network, to reconstruct the scene information from the scanning radar echoes. Unlike the traditional analysis-based method, we established a deep unfolded scene reconstruction network based on the structure of ISTA. This network can learn the optimal network parameters through the input radar data, which avoids the manual selection of parameters in the traditional method. Besides, we apply a loss function to ensure the effectiveness of the sparse transformation so that the method can recover target information from scanning radar echoes in various complex scenes. Extensive experiments demonstrate that this method can highly improve scene reconstruction performance.