YOLO-Based Detection Method for Complex Underwater Acoustic Pulse Signals

In the oceanic environment, complex interference factors such as reverberation, harmonic interference, and multipath effects have severely restricted the performance of traditional underwater acoustic pulse signal detection methods. This study proposes a signal detection method based on YOLOv12, effectively addressing the challenges of detecting underwater acoustic pulses under such complex interference. The process takes the time-frequency spectrogram of underwater acoustic pulse signals as input and incorporates data augmentation, multi-scale transformations, and post-processing tailored to signal characteristics. Subsequently, we leverage the powerful object detection capabilities of YOLOv12 to achieve real-time detection of pulse signals. Simulation results demonstrate that on a synthesized dataset containing reverberation, harmonic, and multipath interference, this method achieves a mean Average Precision (mAP50) of 95.73% and mAP50-90 of 84.69% at a signal-to-noise ratio (SNR) of -10 dB. Compared to traditional energy detection methods, our approach exhibits significant advantages and offers a novel solution for target detection in complex underwater acoustic environments.