부산대학교 도서관

With the increasing global marine pollution, the detection and treatment of marine debris is particularly important. With the continuous development of machine learning, the object detection model has become a powerful tool for marine debris disposal. This paper proposes an improved YOLOvS model [1], [2], which changes backbone to mobileNet[3] on the basis of the original YOLOv5s model and introduces an attention mechanism to filter key features. The results show that on the TrashCan dataset[4], the detection precision and recall rate of our model have reached 79% and 63% respectively, and the detection effect has been improved by 9% and 2% compared with YOLOv5, respectively. Compared with the current underwater target detection model YOLOTrashCan[5], which has higher detection accuracy, the mAP@0.5 has been improved by 2.0%, realizing the accurate detection of marine debris in the real underwater environment. The research shows that our model has high precision while reducing model parameters and improving reasoning speed.