CO_YOLO:基于改进YOLOv5的海洋生物目标检测
CO_YOLO:Marine Biological Object Detection Based on Improved YOLOv5
张俊恒 1司亚超 1郑孟然1
作者信息
- 1. 河北建筑工程学院,河北 张家口 075000
- 折叠
摘要
由于水下环境的复杂性,存在目标重叠和遮挡等情况,为解决这个问题,文章在YOLOv5s的基础上进行改进提出了CO_YOLO,通过将C3替换为C3_ODConv,减少模型的计算量,加入了CBAM注意力机制可以增强模型的特征提取能力,同时.实验结果表明CO_YO-LO对海洋生物目标检测有很好的效果,在URPC2020数据集上比YOLOv5s有更好的效果,mAP50达到0.828,计算量GFLOPs为16.5.
Abstract
Due to the complexity of underwater environment,objects overlap and occluding ex-ist.In order to solve this problem,this paper improved on the basis of YOLOv5s and proposed CO_YOLO.By replacing C3 with C3_ODConv,the computational load of the model was re-duced,and CBAM attention mechanism was added to enhance feature extraction capability of the model.The experimental results show that CO_YOLO has a good effect on Marine biological target detection,and has a better effect than YOLOv5s on URPC2020 data set,with mAP50 reaching 0.828 and GFLOPs reaching 16.5.
关键词
目标检测/海洋生物/C3_ODConv/YOLO/CBAMKey words
Deep learning/Marine life/C3_ODConv/YOLO/CBAM引用本文复制引用
出版年
2024
NETL
NSTL
万方数据