Underwater Object Detection using Image Enhancement and Deep Learning Models

摘要

Autonomous underwater vehicles (AUVs) are efficient robotic tools, offering a wide range of applications in ocean exploration and research, such as oceanographic mapping, environmental monitoring, and archaeology. Incorporating an automatic object detection system with AUVs can substantially improve their ability to perceive and recognize objects in a complicated and often hazardous environment. Currently, detecting underwater objects relied on a man-in-the-loop approach, where AUVs captured vast amounts of data and saved them in memory for offline processing. This study investigates the use of deep learning for automatic image preprocessing and object detection, evaluating and comparing three state-of-the-art YOLO (You Only Look Once) models, including YOLOv8, YOLOv7, and YOLOv5. Extensive experiments were conducted using publicly available underwater image datasets, revealing that the pre-trained models attain superior performance on the Brackish dataset. YOLOv5 and YOLOv8 achieved the highest mean average precision (mAP) with a score of 99%, while YOLOv7 scored 89%. Furthermore, an underwater image enhancement algorithm is employed on the URPC2021 dataset, significantly improving the detection accuracy with a 3% increase in mAP across all three models. In terms of inference speed, YOLOv5 demonstrated the highest frames per second (FPS), while maintaining comparable performance in mAP and recall.