融合多分支网络结构的高频工件图像识别算法  

High Frequency Workpiece Image Recognition Algorithm Based on Fusion of Multi Branch Network Structure

作　　者：孙成龙 李柏林[1] 李节 王逸涵 欧阳 SUN Chenglong;LI Bailin;LI Jie;WANG Yihan;OU Yang(School of Mechanical Engineering,Southwest Jiaotong University,Chengdu 610031,China;School of Mechanical Engineering,Chengdu University,Chengdu 610106,China)

机构地区：[1]西南交通大学机械工程学院,成都610031 [2]成都大学机械工程学院,成都610106

出　　处：《组合机床与自动化加工技术》2024年第7期7-11,共5页Modular Machine Tool & Automatic Manufacturing Technique

基　　金：四川省科技厅重点研发项目(2021YFN0020)。

摘　　要：为有效解决复杂光照变化下高频工件图像识别率低的问题,提出一种融合多分支网络结构的高频工件图像识别算法。该算法以Efficient-b0为基础网络,首先,引入轻量级的混合注意力模块提取强光照鲁棒性的全局工件图像特征,经过主干网络得到全局识别结果;然后,采用弱监督区域检测模块定位工件的局部重要区域,并将其引入分支网络得到局部识别结果;最后,在分支融合模块中联合全局和局部识别结果实现工件识别。实验结果表明,相较于多种图像识别算法,所提出的算法对光照变化具有更强的适应性,显著提高了高频工件识别性能,识别准确率达到了97.8%。In order to effectively solve the problem of low recognition rate of high-frequency workpiece images under complex illumination changes,this paper proposes a high-frequency workpiece image recognition algorithm that integrates multi-branch network structure.The algorithm uses Efficient-b0 as the basic network.Firstly,a lightweight mixed attention module is introduced to extract the global workpiece image features with strong illumination robustness,and the global recognition result is obtained through the backbone network;then,the weakly supervised area detection module is used to locate the workpiece The local important area is introduced into the branch network to obtain the local recognition result;finally,the global and local recognition results are combined in the branch fusion module to realize the artifact recognition.The experimental results show that,compared with various image recognition algorithms,the proposed algorithm has stronger adaptability to illumination changes,significantly improves the performance of high-frequency workpiece recognition,and the recognition accuracy reaches 97.8%.

关 键 词：光照变化 网络结构 混合注意力 区域检测 图像识别 

分 类 号：TH166[机械工程—机械制造及自动化] TG659[金属学及工艺—金属切削加工及机床]