HydroVision: A Comprehensive Approach to Underwater Image Dehazing and Object Detection

Underwater imaging is often challenged by color distortion, haze, and reduced visibility due to the scattering and absorption of light. These issues hinder accurate object detection and classification, posing difficulties for marine exploration, ecological research, and underwater robotics. HydroVision addresses these challenges by combining advanced dehazing, color restoration, and object detection techniques. The first component, Dark Pixel Color Correction, enhances the visibility of underwater images by mitigating backscatter, improving depth range perception, and increasing contrast. Once the images are dehazed, the YOLO (You Only Look Once) object detection model is applied. YOLO was trained on a custom dataset with 15 marine ecosystem classes, including various coral types, marine fauna, and substrates. The integration of Dark Pixel Prior with YOLO significantly improves the clarity and accuracy of object detection, as the dehazed images offer clearer and more distinguishable features. Experimental results demonstrate superior performance across key metrics. HydroVision achieves the lowest Mean Squared Error (MSE), the highest Peak Signal-to-Noise Ratio (PSNR), and an SSIM of 0.904, indicating excellent structural similarity preservation. These advancements position HydroVision as a powerful tool for marine ecosystem monitoring, biodiversity research, and autonomous underwater robotics.