Enhancing Underwater Fauna Monitoring: A Comparative Study on YOLOv4 and YOLOv8 for Real-Time Fish Detection and Tracking

Supervision of marine ecosystems is paramount for understanding and conserving underwater biodiversity. Due to the latest advancements in the field of neural networks and image recognition, computer vision and deep learning techniques have revolutionized wildlife monitoring. YOLO (You Only Look Once) is recognized as one of the most popular and highly accurate computer vision algorithms, designed specifically for object detection. This paper responds to the need for effective underwater fauna monitoring by conducting a thorough comparative study of two cutting-edge state-of-the-art object detection models, YOLOv4 and YOLOv8, in the context of real-time fish detection and tracking within dynamic underwater environments. Performance evaluation for these models encompasses various parameters, including accuracy, real-time processing speeds, and the models’ adaptability to the formidable conditions of underwater environments. For this study, both YOLOv4 and YOLOv8 architectures were trained on a cloud platform. This cloud-based approach circumvents the challenges and logistical complexities of deploying manpower and complex hardware in underwater scenarios, making it a more effective solution for underwater fauna monitoring and yielding superior results. The results obtained substantiate the superiority of the YOLOv8 model over the YOLOv4 model for underwater video surveillance applications, even under different dynamic and challenging conditions. This work underlines the significant role that highly accurate object detection methods like YOLO can play in the quest to safeguard the future of our planet's aquatic life.