Depthwise Separable Convolutional SAR Target Recognition Embedded with Attention Mechanism
The application of Deep Separable Convolution(DSC)makes the deep learning network model lightweight.On this basis,a DSC Syntheic Aperture Radar(SAR)target recognition method embedded with attention mechanism is proposed.By combining the depth separable convolution with the attention mechanism,the ability of network learning the important target features is improved;At the same time,multiple depth separable convolutions are superposed and paralleled,and multi-scale network modules are designed to enhance the feature extraction capability of different depth networks;Finally,residual connection is used to alleviate gradient dispersion and gradient explosion of deep network.Experiments show that the proposed method achieves an average recognition rate of 99.0％under the condition of small network model parameters,and has strong recognition advantages.
万方数据
维普
