A Multi-scale Radar HRRP Target Recognition Method Based on Pyramid Depthwise Separable Convolution Network

Abstract

Radar high resolution range profile (HRRP) contains important structural features such as target size and scattering center distribution, which has attracted extensive attention in the field of radar target recognition. In order to solve the problem of feature extraction and recognition in HRRP target recognition, we propose an HRRP target recognition method based on one-dimensional Pyramid Depthwise Separable Convolutional (PyDSC) neural network. For the processed data, pyramid convolution is selected, and convolution kernels of different sizes are used on different input channels, which can better extract the features of different scales and improve the overall recognition ability. At the same time, Depthwise Separable Convolution (DSC) technology is applied to PyConv network, a standard convolution operation is divided into two steps: deep convolution and point convolution, which can reduce the network complexity, reduce the amount of parameters and improve the speed of HRRP target recognition. Finally, we verify the effectiveness of the proposed method through experiments. The experimental results show that: 1) compared with the other three convolutional neural networks, our proposed PyDSC can significantly improve the recognition accuracy with a small increase in overhead; 2) Compared with the original PyConv, PyDSC can effectively reduce the complexity of the model.