An Enhanced Algorithm for Marine Litter Detection in YOLOv5

The ocean is a crucial component of Earth's ecosystem; however, marine debris pollution poses a significant threat to the integrity of this system. Detecting marine debris is imperative for autonomous garbage detection and removal, and currently, target detection based on the YOLOv5 network is widely adopted. Firstly, owing to challenging lighting conditions and issues related to partial target occlusion in the marine environment, this paper introduces the ECA attention mechanism to enhance the model's focus on crucial areas. Secondly, to optimize the model for deployment, the study evaluates the computational complexity of the network with the incorporation of the DSConv lightweight network. Finally, to strike a better balance between detection performance and false positives, the Wise-Iou loss function is introduced. Ultimately, experimental validation on the Trash_Icra19 dataset demonstrates a 2.6% improvement in detection accuracy and a 12.5% reduction in computational complexity for the enhanced model.