Underwater Biological Target Detection Algorithm Based on MSD-YOLOv5

Abstract

In oceanography, target detection technologies play a key role in various applications, such as assisting underwater robots in detecting the seabed, identifying ships and people at sea, and helping aquaculture farmers monitor the types and distribution of aquatic organisms, adjusting culture densities to ensure their growth and development. However, challenges like small aquatic target size, overlap and occlusion issues, poor underwater lighting conditions, and high ambient noise and interference can result in reduced detection accuracy. Moreover, real-time detection is essential in aquaculture. To address these issues, this paper proposes the MSD-YOLOv5 model, which integrates Space-to-Depth into the Backbone and Neck of YOLOv5 and replaces the Head of YOLOv5 with a separated Head to enhance model accuracy while meeting real-time detection requirements. Experimental results demonstrate that the MSDYOLOv5 model improves [email protected] and [email protected]:0.95 by 2.7 percentage points compared to the YOLOv5 model, and achieves a detection speed of 196 FPS, meeting the requirements for real-time detection of underwater organisms.