The intelligent object detection framework for detecting fish from underwater images

Marine applications heavily rely on underwater object detection, yet challenges like complex backgrounds and image quality issues impede deep learning-based detectors. Monitoring feed pellet utilisation in aquaculture is vital for efficient resource management. This study introduces a novel framework, DYNFS, merging underwater object detection and image reconstruction using YOLO-V5. Initially, we curate an underwater image dataset, refining it to remove noise, and then employ DYNFS for classification. Our approach achieves a 98.93% accuracy rate in identifying submerged feed pellets, crucial for aquaculture efficiency. However, locating pellets remains challenging due to poor image quality and small object sizes. The enhanced YOLO-V5 networks show promise in real-world aquaculture scenarios. This framework enhances underwater object detection, offering potential benefits for marine applications and aquaculture management.