Efficient multi-perspective jamming feature perception network for suppressive jamming recognition with limited training samples

Abstract

Recognising suppressive jamming signals is crucial for radar systems to counteract this type of jamming, highlighting the importance of research in this area. Current deep learning-based methods for identifying suppressive jamming signals suffer from reduced effectiveness with limited training samples and issues related to high parameter counts and computational complexity. To address these challenges, the authors propose a jamming recognition method based on an efficient multi-perspective jamming feature perception network. This method extracts features from the time-frequency spectrum of jamming signals from multiple perspectives, including local, multi-scale, cross-space, and global, to obtain more robust and discriminative jamming features and improve recognition under limited training sample conditions. Additionally, the authors design efficient modules for local jamming feature extraction, multi-scale jamming feature down-sampling, and global jamming feature representation. The lightweight design of these modules enables the proposed method to maintain excellent jamming recognition performance while reducing parameters and computational complexity. Simulation experiment outcomes highlight the exceptional effectiveness of the proposed technique across multiple metrics compared to eight other approaches. Furthermore, the proposed method exhibits significantly fewer parameters and lower computational complexity than its deep learning-based counterparts.