Radar Compound Deception Jamming Recognition Based on Fast–Slow Time–Frequency Distributions

In complex countermeasures environments, the recognition of the type of jamming is an important prerequisite for radar to accurately measure target information. The time–frequency distribution (TFD) image is among the most prevalently used features for this purpose. Nevertheless, in the background of compound deception jamming (CDJ), the intertwining of diverse jamming types in conventional TFD images is severe, which leads easily to missed detection and erroneous recognition of small-size jamming, especially in scenarios with a low jamming-to-noise ratio (JNR). Hence, this article presents a CDJ recognition method based on fast–slow TFD (FST-FD) to improve the reliability of radar measurements. This approach integrates coherent accumulation (CA) to enhance recognition performance under low JNR conditions and capitalizes on the differences in Doppler frequency shifts to mitigate the feature interweaving among different jamming types in FST-FD images. Specifically, this article first validates the transfer invariance of TFD information during the CA process. As a result, the FST-FD information of CDJ is effectively separated and extracted in the range–Doppler domain. Subsequently, to boost the recognition performance for jamming types with small FST-FD areas, a mixed attention mechanism-YOLOv8n (MAM-YOLOv8n) lightweight network incorporating both channel and spatial attention mechanisms is used, which not only reduces missed detections but also ensures real-time performance for CDJ recognition. Compared with published recognition methods, our method obtains 2.6%–8% and 12.1%–16.9% improvement in recognition accuracy on the public measurement dataset and on our measurement dataset, respectively. This advantage holds significant implications as it enables the recognition of sidelobe or long-distance jamming with low JNRs.