轻量化YOLOv5s的水下垃圾检测方法

打开文本图片集

中图分类号：TP391.41 文献标志码：A

DOI：10. 13338/j. issn.1674-649x.2025.02.005

Underwater garbage detection method using lightweight YOLOv5s

WANG Yannian ，LI Ying，LIAN Jihong （School of Electronics and Information，Xi'an Polytechnic University，Xi'an 71oo48，China）

Abstract Aimed at the problems of low recognition accuracy and low speed of smalltarget detection algorithm，an improved lightweight YOLOv5s marine underwater garbage detection method， namely YOLOV5s-MCS，was proposed. Although the backbone network of YOLOv5s itself had a strong feature extraction capability，in the face of small underwater garbage fragments，the continuous deepening of convolution might lead to the loss of some features and similar redundancy between feature graphs.Therefore，in order to minimize the number of parameters and improve the speed of the model，MobileNetv3-Small was proposed to optimize the backbone network in YOLOv5s.Secondly，coordinate attention （CA） was used to replace the original atention mechanism SE in MobileNetv3-Small.In this way，not only the information between channels，but also the position information of horizontal and vertical coordinates could be obtained. Finally，the CIoU loss function was optimized to SIoU loss function.By improving the overall network model，the detection accuracy reaches 85.7% while ensuring the model's lightweight，and the number of floating-point operations and parameters are reduced to 1/5 and 1/7 of the benchmark YOLOv5s network，respectively.

Keywords marine underwater garbage detection; YOLOv5s; lightweight; attention mechanism; loss function

0引言

据科学家估计，目前海洋中存在 万吨的塑料垃圾，其中绝大部分垃圾都是位于深海，这些塑料的降解需要百年左右，然而由于光线、温度等原因，深海垃圾降解的时间需要更久。（剩余13248字）