Marine Radar Small Target Classification Based on Normalized Time-Frequency Image and Residual Network

In this article, a small target classification method is proposed, which can effectively distinguish different types of sea surface targets under different sea states and different radars. The proposed classification method is mainly based on normalized time-frequency spectrograms and Residual Network (ResNet). The time-frequency images of the radar returns reflect the motion characteristics of the target, the fluctuation of the target Radar Cross Section (RCS) and the relationship between the target and the wave. The effect of different radar parameters on target classification can be eliminated by standardizing range resolution and carrier frequency of radar returns. By exploiting ResNet, the proposed method can effectively extract the deeply hidden features of normalized time-frequency images and solve the problems of gradient disappearance, gradient explosion and degradation in deep network. Datasets consisting of four types of targets, floating ball, floating boat, yacht and low altitude Unmanned Aerial Vehicle (UAV) are constructed to test the performance of the proposed method. The experiment results show that the proposed classifier combines normalized time-frequency images and ResNet attains better classification performance compared with the existing classification methods.