Real-Time AI-Based Detection of Fish and Humans in Underwater Environments

Underwater visual monitoring is difficult because light attenuation, turbidity, color distortion, occlusion, and changing backgrounds can reduce the reliability of conventional object detection. This work presents an underwater human and fish detection system that combines a waterproof camera, image preprocessing, YOLO-based real-time object detection, and convolutional neural network (CNN) classification. The proposed system is intended to detect humans and fish in underwater video, classify fish by species and gender, and present the results through a monitoring dashboard. Human status is considered using observed movement and posture, while fish classification uses labeled underwater data such as the Fish4Knowledge dataset described in the project study. The implementation uses OpenCV and Python-based deep-learning components and provides real-time visualization with bounding boxes and class labels. The project report also defines tests for clear and low-visibility water, multiple objects, fast motion, partial occlusion, reflections, varying depth, debris, overlapping objects, and multi-frame operation. The demonstrated system provides a foundation for underwater surveillance, marine research, biodiversity monitoring, and search-and-rescue applications.