A Novel Long-Tailed Aerial Target Classification Method Based on Adaptive Decoupled Learning

Radar Automatic Target Recognition (RATR) is important in modern radar technology. Classifying aerial targets using Micro-Doppler (M-D) features extracted from narrowband echoes matters for ground personnel's decision-making. Deep learning (DL) is widely used in M-D recognition but faces challenges due to long-tailed distribution of aerial target data in engineering. This paper proposes a novel long-tailed (LT) micro-Doppler recognition method with adaptive decoupled learning (ADL), separating learning into feature extraction and classification stages. It's tested on a real dataset and a signal processing procedure for real data is developed. Experimental results show the proposed method performs better than classical DL M-D recognition and other long-tail identification methods.