Automatic Extraction of Bearing Trajectory of Underwater Acoustic Sources Based on Bearing Time Recording

Bearing Time Recording (BTR) map constructed by the spatial spectra of multi-frame data is often applied for source detection in passive sonar systems. Automatically extracting bearing trajectory of underwater acoustic sources in BTR map can make passive detection technology applicable to underwater unmanned platforms. However, owning to the difficulties of weak targets detection and multi-target resolution, it is a challenge for automatic detection of underwater targets. Considering the advantages of YOLOv8 in speed and accuracy, this article applies the object detection network YOLOv8 to bearing trajectory extraction in BTR map. We create BTR dataset using acoustic data collected in South China Sea and finetune the YOLOv8 network on the BTR dataset. The retrained YOLOv8 model is used to detect target regions in BTR maps. Then we map every target region to a numerical matrix and extract source bearing trajectory by finding the maximum value of each row within the matrix. The results show that YOLOv8 can basically achieve object detection in BTR maps and bearing trajectories are precisely extracted by finding the local maximum value of the numerical matrix of BTR maps.