基于改进YOLOv5的轴承座表面缺陷模型检测  

Bearing Housing Surface Defect Model Detection Based on Improved YOLOv5

作　　者：梁世金 杨旗[1] LIANG Shi-jin;YANG Qi(School of Mechanical Engineering,Shenyang Ligong University,Shenyang 110159,China)

机构地区：[1]沈阳理工大学机械工程学院,辽宁沈阳110159

出　　处：《机械工程与自动化》2024年第4期134-137,共4页Mechanical Engineering & Automation

摘　　要：针对零件(以轴承座为例)表面缺陷人工检测效率低、容易误检和漏检的问题,提出了一种基于YOLOv5的改进的检测算法。首先在主干网络中引入CA注意力机制,该机制将位置信息嵌入到了通道注意力中,有效提升了模型的检测性能。其次,将YOLOv5的定位损失函数由CIoU改为WIoU v3 loss,以提高预测框回归精度。接着用轻量级卷积方法GSConv代替原有的标准卷积,使网络模型计算量及参数量降低,以提升模型推理速度。实验结果表明,所改进算法与YOLOv5s原模型相比,参数量减少了6.1%,计算量减少了3%,平均检测精度提升了1.3%,检测速度提升了1.6%。Aiming at the problem that the manual detection efficiency of surface defects of parts(taking bearing housing as an example)is low,and it is easy to misdetect and miss detection,this paper proposes an improved detection algorithm based on YOLOv5.Firstly,the CA attention mechanism is introduced in the backbone network,which embeds the position information into the channel attention to effectively improve the detection performance of the model.Secondly,the location loss function of YOLOv5 is changed from CIoU to WIoU v3 loss to improve the regression accuracy of prediction box.Then,the lightweight convolution method GSConv is used to replace the original standard convolution,which reduces the computation and parameter amount of the network model and improves the model inference speed.Experimental results show that compared with the original YOLOv5s model,the improved algorithm reduces the number of parameters by 6.1%,the amount of computation decreases by 3%,the average detection accuracy is increased by 1.3%,and the detection speed is increased by 1.6%.

关 键 词：目标检测 注意力机制 轻量化 深度学习 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]