A CNN and YOLOv5-Based Deep Learning Pipeline for Underwater Image Enhancement, Object Detection and Classification

Underwater image processing and computer vision are difficult because of low illumination, colour distortion, and limited view. This work introduces a deep learning pipeline using CNNs and YOLOv5 for object recognition, categorization, and underwater picture augmentation. Along with identifying and categorizing marine creatures and structures for scientific study, conservation, and environmental monitoring, the pipeline improves underwater photography. A CNN-based design removes poor contrast, noise, and color shift during the picture enhancing phase. Trained on a variety of underwater images, the CNN model enhances characteristics and visibility to translate degraded input images into more realistically visible outputs. For object detection, the updated picture is clearer and more beneficial. Items are identified and classified using YOLOv5, a state-of- the cutting edge real-time detection system. Underwater real-time uses would find YOLOv5 perfect given its speed and precision. A special collection of annotated underwater photos of objects and animals enhances the accuracy and resilience of the model in many underwater environments. After training, YOLOv5 very accurately identifies marine species, fish, and other underwater objects. Two main components and a post-processing step utilizing a confidence threshold filter to confirm detection findings define the pipeline. This guarantees that under close examination only correct forecasts are shown. Appropriate for field deployment in AUVs or ROVs, the suggested system is light-weight and driven towards real-time processing.