Open Set Recognition of Radar Signals Based on Time-Frequency Fusion and OVA Network
Xinqiao Yang, Dongqing Zhou, Qi Zhang, Xin Ge, Xianhua Wang, Chunxi Dong, Yang‐Yang Dong · 2023
For open electromagnetic environment, most of the radar signal modulation recognition methods available cannot handle the open set recognition well. To overcome this problem, a time frequency fusion and One-vs-All (OVA) network based radar signal open set recognition algorithm is proposed. Firstly, combining the advantages of Wigner-Ville distribution (WVD) with high time-frequency resolution and Choi Williams distribution (CWD) with a good cross terms suppression ability, a multi time-frequency image feature fusion algorithm based on wavelet transform is proposed, where fusion criteria for low-frequency and high-frequency components of the image are designed to improve the quality of time-frequency images. Then, a radar signal open set recognition method based on OVA network is proposed, where a closed set classifier is used for pre-classification, and the OVA network is trained with the most similar negative samples to maximize the classification boundary of similar categories, and its classification confidence is improved via open set entropy minimization strategy and threshold judgment. For closed set scenario simulations, it can achieve 92.6% accuracy with signal to noise ratio (SNR) of -6 dB and have the ability to handle small sample cases. What's more, for open set scenarios, both known and unknown modulation types of radar signals can be recognized well.