Physical IoU-Based YOLO Network for Shortwave Communication Burst Signal Recognition

Shortwave communication plays a critical role in military and emergency applications. However, the complex nature of shortwave transmission channels makes automatic burst signal recognition particularly challenging, especially in non-cooperative scenarios. This letter presents an optimized recognition method based on the You Only Look Once (YOLO) network, which accurately identifies signal types while improving the precision of burst time detection. A Physical Intersection over Union (PIoU) is proposed by leveraging the bounding box’s left edge, corresponding to signal burst time, to refine bounding box regression and resolve ambiguities in conventional IoUs. Experiments show the proposed PIoU improves the accuracy of burst time detection by 10.44% under Signal-to-Noise Ratios (SNRs) from -20dB to 18dB, compared to the state-of-the-art IoU methods. Further tests across 8 shortwave communication channels confirm consistent improvements of burst time detection, while other metrics such as Precision, Recall, mean Average Precision at 50% IoU(mAP50), and mAP95 are maintained. These advancements significantly benefit signal decoding and information restoration.