Recognition of SAR Active Compound Jamming via Multidomain Multilabel Learning

With the rapid development of electronic warfare technology, numerous complex scenarios involving active compound jamming pose significant challenges to Synthetic Aperture Radar (SAR) systems. Recognition of jamming patterns in such environments is crucial for effective anti-jamming strategies. However, existing algorithms suffer from issues such as low accuracy in the compound jamming, the high network complexity when handling a large number of jamming labels, and insufficient utilization of features in SAR echo data. These issues make the algorithms difficult to apply effectively to SAR instruments under low jamming-to-signal ratio (JSR). This article proposes a multi-domain multi-label learning-based SAR compound jamming pattern recognition algorithm to solve these problems. First, to make use of the correlation and complementarity between the multi-domain information in SAR echo data, the time domain and time-frequency domain features are fused to realize a more effective and comprehensive expression of features. Then, a dual-branch ShuffleNet V2 architecture is designed, where the convolutional block attention module (CBAM) attention mechanism is introduced, and a multi-scale convolutional module is designed to significantly enhance recognition capability. Finally, independent multi-label discrimination thresholds are generated using a particle swarm optimization algorithm to improve the recognition performance. The experiment shows that the proposed method outperforms existing classical recognition methods in terms of recognition performance. When the JSR is 2 dB, the overall recognition accuracy of this algorithm can reach 92.71%, which represents a relative improvement of 2.43% ∼ 25.45% compared with existing methods.