Prototype Features Driven High-Performance Few-Shot Radar Active Jamming Recognition

Abstract

Accurate identification of jamming is the premise of effective work of radar anti-jamming systems. As the electromagnetic environment becomes increasingly complex, radar detection faces not only the issue of insufficient training samples but also the challenge of imbalanced jamming samples. To solve this problem, this article proposes a few-shot recognition method of radar active jamming guided by prototype features. In this method, a pyramid structure is used to construct feature maps at different levels to integrate low-level features and high-level semantic features, so as to retain the information of the time-frequency images of the jamming signal to the maximum extent. Meanwhile, a differentiation information attention module is introduced to capture the global and local information of the feature maps and enhance the signal perception ability of the model. Finally, we propose a prototype feature extraction and fusion module to learn the prototype features of various samples and fuse them with backbone features. In view of the uneven data of the training set, the imbalanced coefficient is proposed to improve the recognition accuracy of the few-shot jamming signal in a complex electromagnetic environment. The experimental results on the jamming simulation dataset show that the proposed model has good recognition accuracy and robustness, and can handle imbalanced jamming samples. When the jamming-to-noise ratio (JNR) exceeds 2 dB, the average recognition accuracy of jamming can reach 99%. In the case of low JNR and sample imbalance, the proposed structure can effectively identify multiple small classes of jamming.