Deep Learning Pruning Based SAR Ship Target Recognition

In this paper, a method based on sparse pruning of deep learning channels is proposed to address the problems of large number of network parameters and computational complexity during the recognition of ship targets in SAR images. By applying L1 regularization on the scaling factor of the batch normalization (BN) layer, the identification and pruning of unimportant channels in the CNN is achieved, which reduces the number of parameters and computational complexity of the network. A series of lightweight networks are generated by pruning the VGG-19 network with channel sparsification. By comparison, it is demonstrated that the proposed channel pruning method has a significant advantage in reducing the number of model parameters, and also performs well in the recognition rate.