Classification of Space Targets with Micro-motion Based on Deep CNN

Abstract

There exist a variety of micro-motions in space targets, including nutation, precession and spinning. Accurate acquisition of the micro-motion form is a prerequisite for estimating motion and structure parameters of ballistic targets. Firstly, we analyze the micro-Doppler representations under three kinds of micro-motion forms, and the time-frequency maps of radar echo signal are generated as the data set. Then we retrain AlexNet and SqueezeNet using transfer learning to classify the micro-motion form. We also study the effect of noise on the classification performance. Simulation results show the effectiveness of the proposed method, which provides an instructive value for the space target recognition.