改进YOLOv5s的轻量化零件目标检测算法  

Improved Target Detection Algorithm for Lightweight Parts of YOLOv5s

作　　者：孙家慧 王赫莹[1] 郭忠峰[1] SUN Jiahui;WANG Heying;GUO Zhongfeng(School of Mechanical Engineering,Shenyang University of Technology,Shenyang 110000,China)

机构地区：[1]沈阳工业大学机械工程学院,沈阳110000

出　　处：《组合机床与自动化加工技术》2024年第11期135-141,共7页Modular Machine Tool & Automatic Manufacturing Technique

基　　金：辽宁省教育厅2021年度科学研究经费项目(LJKZ0114)。

摘　　要：针对无序密集放置的多尺度零件识别难度大等问题,提出一种基于改进YOLOv5s的轻量化零件目标检测算法。在Backbone网络C3模块添加CA注意力机制,提高目标特征提取能力;引入了damo-yolo的Efficient-RepGFPN结构代替原有的Neck层,以减轻模型的复杂度;引入SimAM注意力机制,提高卷积网络的表征能力;为加快计算速度,降低运算成本,用轻量级卷积GSConv代替Neck结构中的标准卷积;采用FocalEIOU替换YOLOv5算法中的CIOU对模型识别性能进行优化。实验结果表明,在自制零件数据集上,改进算法的mAP@0.5达到99.4%,检测速度仅需5.7 ms,FPS达到175帧/s,且计算量和参数量都大幅度降低,模型大小仅有原来的32%,易于移动端部署,在零件检测精度、检测速度等方面均优于原有YOLOv5s,满足视觉引导下对零件精准识别。In order to solve the problem of difficulty in identifying multi-scale parts with disordered and dense placement,this paper proposes a target detection algorithm for lightweight parts based on improved YOLOv5s.The CA attention mechanism is added to the C3 module of the backbone network to improve the target feature extraction ability.The Efficient-RepGFPN structure of damo-yolo was introduced to replace the original Neck layer to reduce the complexity of the model.The SimAM attention mechanism is introduced to improve the representation ability of convolutional networks.In order to speed up the computation speed and reduce the computational cost,the lightweight convolution GSConv is used to replace the standard convolution in the Neck structure.FocalEIOU was used to replace the CIOU in the YOLOv5 algorithm to optimize the recognition performance of the model.The experimental results show that on the self-made parts dataset,the mAP@0.5 of the improved algorithm reaches 99.4%,the detection speed is only 5.7 ms,the FPS reaches 175 frames/s,and the amount of calculation and parameters are greatly reduced,the model size is only 32%of the original,easy to deploy on the mobile terminal,and is better than the original YOLOv5s in terms of part detection accuracy and detection speed,so as to meet the accurate identification of parts under visual guidance.

关 键 词：YOLOv5s 注意力机制 Efficient-RepGFPN SimAM注意力机制 轻量级卷积 

分 类 号：TH16[机械工程—机械制造及自动化] TG506[金属学及工艺—金属切削加工及机床]