Underwater Image Enhancement and Restoration with YOLO-based Object Detection and Recognition

Abstract

Underwater image quality can suffer greatly due to light absorption, light scattering, color distortion, and noise, which affect image quality and object detection capabilities in marine research, autonomous navigation, and underwater exploration. In this study, we present a novel, improved YOLO-based framework for real-time underwater object detection and image enhancement in the presence of these difficulty layers. The proposed system consists of three main modules: the image enhancement module, which includes adaptive histogram equalization to restore image quality; the segmentation module, which has a modified YOLO backbone and new loss functions to produce pixel-level masks; and the detection module for refining bounding boxes and removing false positives. The paper also measures and evaluates the behavioural performance of YOLOv10 with YOLOv13 models through evaluation metrics of precision, recall, F1 score, and mAP50 for object detection performance in challenging underwater environments. With reference to detecting objects underwater, the work shows YOLOv13 is successful and has a precision of 99.2%, recall of 98.8%, and F1 score of 98.5%. The study recommends that improved photo-enhancing techniques are able to improve real-time processing, performance, and frames per second in small resource-constrained applications, like underwater robotics. This method has potential for use in commercial, conservation, and underwater research.