Underwater shipwreck target detection based on audio-visual heterogeneous data

As marine exploration continues to grow, the frequency of maritime emergency contingencies increases year by year. It is crucial to enhance marine surveillance to effectively detect underwater maritime shipwreck targets. Based on the excellent target detection performance of the YOLO5 model, we constructed the pure visual image dataset, pure sonar image dataset, and audio-visual heterogeneous dataset. We also trained the YOLOv5m model with a personal computer using the visual image dataset, sonar image dataset, and the hybrid audio-visual image dataset of maritime shipwrecks respectively. The test results show that the detection accuracy of the model trained with the hybrid maritime wrecks audio-visual image dataset is higher than that of the model trained with pure visual and sonar image datasets. This method may have an inspiration to future research.