YOLOv5s-GTB: An Improved Underwater Garbage Detection Algorithm Based on YOLOv5s

To address the issues of misjudgment, omission, and high computational complexity in underwater garbage detection, this paper proposes the YOLOv5s-GTB underwater garbage dection algorithm based on YOLOv5s. Firstly, to make the model lightweight, this paper proposes introducing the GhostBottleneckV2 module in the Neck. Secondly, The Transformer module is utilized in the final layer of the Backbone to enhance the feature extraction capability of the network. Finally, the BiFPN concept is implemented in the Neck, incorporating skip connections to enrich semantic feature representation and enhance detection performance. The experimental results demonstrate that compared to the original model, the improved model increases mAP by 0.9%, reduces parameters by 11.9%, decreases model size by 11.0%, and reduces GELOPS by 10.7%. The result shows YOLOv5s-GTB has better performance.