Enhanced Underwater Object Detection Using YOLO Models for Robust Marine Ecosystem Monitoring

Underwater object detection is challenged by low contrast, poor visibility, and inconsistent lighting, limiting traditional algorithms. This study addresses the issues using YOLO models known for their exceptional effectiveness in identifying marine life in submerged settings. A new dataset has introduced comprising 4,005 annotated images, covering 12 underwater object classes, such as fish, jellyfish, starfish, was created by combining two public datasets to enhance model robustness. Preprocessing with CLAHE improved image quality by enhancing contrast and reducing noise. The dataset was used to evaluate YOLO models, including YOLOv8, YOLOv9, and YOLOv10, in which YOLOv8 with CLAHE achieved the highest detection accuracy. The evaluation utilized standard metrics such as mean Average Precision (mAP), precision, and recall, highlighting YOLOv8's superior performance in underwater object detection with variable lighting, and cluttered scenes.