SAR image recognition method of improved convolutional neural network
SAR images exist speckle noise and distinction between different image categories is poor,which makes it difficult to extract representative features.In order to solve this problem,an improved convolutional neu-ral network for SAR image recognition method is proposed.Firstly,a multi-scale feature extraction module is de-signed to fully extract the hidden information of SAR images by using convolution layers of different scales.Then,by improving classical residual neural network residual block,a dense residual block structure is designed to provide rich detailed information for the back layer and ensure the expression ability of output features.Finally,verification is performed on the MSTAR dataset.The experimental results show that the recognition rate of the proposed model on the test set is 99.17%,which is better than other methods.After salt-pepper noise of different proportions is added to the test set,the proposed model still shows good robustness.
国家科技期刊平台
NSTL
万方数据
