A Lightweight Sea Urchin Detection Algorithm Based on Improved Yolov8

Abstract

In complex underwater environments with rocky reefs or gravelly substrates, an efficient sea urchin and starfish detection method is critical to replace traditional artificial inspection methods. This paper presents a sea urchin and starfish detection method based on an enhanced YOLOv8 (You Only Look Once) model. The backbone and head networks of the model utilize C3Ghost to reduce model size and computation, and an EMA attention mechanism is employed to improve recognition accuracy. The results show that the proposed Yolov8n-GNA model achieves an accuracy of 86.7%, an 11.3% improvement over the original Yolov8n model while maintaining similar recall and mean Average Precision (mAP) values. Furthermore, the computational effort is reduced to 5.4 GFLOPS, representing a 33% reduction compared to the original model. The Yolov8n-GNA model can rapidly and accurately detect sea urchins and starfish in challenging underwater environments, making it suitable for sea urchin aquaculture and fishing operations in complex marine habitats.