SAR image target recognition algorithm based on improved residual shrinkage network

Abstract

Target recognition in SAR image has always been a research hotspot in the world. Aiming at the problem of low target recognition rate in SAR image, this paper proposes a neural network model suitable for SAR image classification, improves the residual shrinkage network, and uses two-channel one-dimensional convolution to improve the residual shrinkage network. On the premise of consuming only a small amount of computation, the information extraction degree of the module is improved, and it is used as the backbone to build the model. On the premise of low parameter quantity and complexity, the recognition rate is 99.4%.