POLSAR Vehicle Target Recognition Based on Complex-Valued Non-local ResNet

Following the superior performance of complex-valued (CV) neural network in the terrain classification of polarimetric synthetic aperture radar (POLSAR) images in recent years, this paper proposes a CV non-local residual network (CV-NL-ResNet) for POLSAR vehicle target recognition, which utilizes both the amplitude and phase information of complex POLSAR data. In the proposed network, CV residual blocks are employed to extract deeper target features. Moreover, a proposed CV version of non-local block is applied to enable CV-NL-ResNet to capture long-range dependencies of the images and pay more attention to the target areas, so as to further improve the accuracy of target recognition. Experiments based on the GOTCHA dataset verify the effectiveness and superiority of the proposed method. The experimental results confirm that CV-NL-ResNet can perform better on POLSAR vehicle target recognition task, compared with the corresponding real-valued (RV) network and other existing well-performing CV networks.