基于改进YOLOv5s的海洋垃圾目标检测算法

Marine garbage target detection algorithm based on improved YOLOv5s

为了减少海洋垃圾对生物和水资源造成的危害,提出一种基于改进YOLOv5s的海洋垃圾目标检测算法。针对目标检测算法中小目标漏检及特征提取能力不足等问题,在主干网络部分添加一种基于大核注意力(LKA)机制的改进模块对特征层进行关键目标提取,并改变了空间金字塔池化(SPPF)网络结构,目的是让深层特征图上的每个像素点在输入图像上映射的区域变大。实验结果表明,改进后的YOLOv5s目标检测模型的平均精度均值(mAP)提升6%。

In order to reduce the harm caused by marine garbage to organisms and water resources,a marine garbage target detection algorithm based on improved YOLOv5s is proposed.To address the issues of missing small targets and insufficient feature extraction capabilities in target detection algorithms,an improved module based on the large kernel attention(LKA)mechanism is added to the backbone network to extract key targets from the feature layer,and the SPPF network structure is changed to make the area mapped on the input image by each pixel of the deep feature map larger.The experimental results show that the mean average precision(mAP)of the improved YOLOv5s target detection model increases by 6%,indicating that the improved algorithm is effective and feasible in the application of marine garbage target detection.