基于改进YOLOv8s 的水下目标检测方法  

Underwater Object Detection Method Based on Improved YOLOv8s

作　　者：冯迎宾 符珊 FENG Yingbin;FU Shan(Shenyang Ligong University,Shenyang 110159,China)

机构地区：[1]沈阳理工大学自动化与电气工程学院,沈阳110159

出　　处：《沈阳理工大学学报》2025年第3期7-15,23,共10页Journal of Shenyang Ligong University

基　　金：辽宁省教育厅高等学校基本科研项目(LJKMZ20220614)。

摘　　要：针对水下环境存在的色偏、细节模糊以及水下目标存在多尺度和遮挡等问题,提出了一种改进YOLOv8s的水下目标检测算法SLG-YOLOv8s。首先,通过Shallow-UWnet网络对水下图像进行增强,提高图像的对比度和清晰度;其次,提出了轻量化多尺度全局注意力(lightweight multi-scale global attention,LMGA)模块,将此模块与YOLOv8s主千部分的C2f融合,通过动态调整权重进行特征重标定,加强特征表达能力,减少计算量;最后,通过聚集和分发(gather-and-distribute,GD)机制采用跨层特征融合的方式增强中间层的信息融合能力,从而提升模型对多尺度目标的检测效率。实验结果表明,SLG-YOLOv8s算法的平均精度均值达到了95.1%,与YOLOv8s算法相比平均精度均值提升了5.1%,精确率和召回率分别提升了4.5%和5.2%,IoU为0.5时海参、海胆、海星、扇贝检测的平均精确率分别提升了4.3%、5.3%、5.6%、5.5%,研究结果可为水下机器人精准捕捞提供重要依据。To address the issues of color bias,blurring of details,and multi-scale and occlusion of targets in the underwater environment,an improved underwater target detection algorithm SLGYOLOv8s is proposed,which improves YOLOv8s.Firstly,the underwater image is enhanced through the Shallow-UWnet network to improve the contrast and clarity of the image.Secondly,a lightweight multi-scale global attention(LMGA)module is proposed,which integrates this module with the C2f of the YOLOv8s backbone.By dynamically adjusting weights for feature recalibration,the feature expression ability is enhanced and the computational workload is reduced.Finally,the information fusion capability of the intermediate layer is enhanced through the Gather and Distribute(GD)mechanism using cross layer feature fusion,thereby improving the detection efficiency of the model for multi-scale targets.The experimental results show that the mAP of the SLG-YOLOv8s algorithm can reach 95.1%,which is 5.1%higher than the YOLOv8s algorithm.The precision and recall rates are improved by 4.5%and 5.2%,respectively.When the IoU of sea cucumber,sea urchin,sea star,and scallop is 0.5,the average precision increases by 4.3%,5.3%,5.6%,and 5.5%,respectively,and the results of the study can be much helpful for underwater robot precision fishing.

关 键 词：水下目标检测 YOLOv8s Shallow-UWnet 跨层特征融合 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]