CNN Based Classi.cation of Rigid Targets in Space Using Radar Micro-Doppler Signatures

Abstract

Micro-motion characteristics play an important role in some applications of radar target classi.cation. In this paper, a classi.cation method of rigid targets in space using radar micro-Doppler signatures is proposed. Based on the attitude kinematics of rigid targets, we analyze feasibility of classi.cation using micro-Doppler signatures by the relationship among inertial properties of typical rigid targets, their micro-motion characteristics, and corresponding modulation to radar echoes. According to the micro-Doppler time-frequency distribution of echoes and the scale of training sample set, Convolutional neural network (CNN) based feature extraction method and softmax Classi.er are designed. Simulations are carried out to validate its e.ectiveness and discuss the impact of observation duration, composition of training data and size of convolutional kernels on its classi.cation robustness and computational cost.