Aerial Target classification using micro-Doppler spectrum based on PCT and ResNet34

The application of deep neural networks has dramatically improved the accuracy of Radar Automatic Target Recognition (RATR). In most involved studies, however, the recognition method based on micro-Doppler (m-D) features has not considered the effects of real noise and clutter,making it difficult to apply in paractical engineering. In this paper, a robust aerial classification method using m-D signatures based on Polynomial Chirplet Transform (PCT) and ResNet is proposed. The Doppler spectrum of the signal is extracted after PCT processing and then classified via ResNet Experiments show that PCT processing improves the differentiability of the Doppler spectrum, and the accuracy of PCT-ResNet is higher than that of ResNet and other machine learning methods.