基于YOLOv5s的晶粒检测算法  

Wafer Detection Algorithm Based on YOLOv5s

作　　者：张子优 韩华超[2] 魏东[1] 于霞[1] ZHANG Ziyou;HAN Huachao;WEI Dong;YU Xia(School of Information Science and Engineering,Shenyang University of Technology;Shenyang Academy of Instrumentation Science CO.,LTD.)

机构地区：[1]沈阳工业大学信息科学与工程学院 [2]沈阳仪表科学研究院有限公司

出　　处：《仪表技术与传感器》2024年第1期114-118,共5页Instrument Technique and Sensor

摘　　要：为解决传统晶粒定位容易受到光线、噪声等因素影响,耗费大量资源等问题,构建晶粒检测算法YOLOv5s-wafer。首先建立晶粒检测数据集,使用轻量级网络GhostNetv2作为主干特征提取网络,降低模型参数量;其次在特征融合网络中引入CA(coordinate attention,CA)注意力机制,加强特征提取能力;最后采用EIOU作为定位损失函数,提高晶粒检测精度。实验结果表明:算法的平均精度均值为99.3%,参数量为4.637×10^(6),检测性能和算法轻量化达到了理想平衡。In order to solve the problem that traditional wafer positioning is easily affected by lightness,noise and other factors and consume a lot of resources,the wafer detection algorithm YOLOv5s-wafer was constructed.The wafer detection dataset was constructed firstly,the lightweight network GhostNetv2 was used as the backbone feature extraction network to reduce the amount of model parameters.Then the CA attention mechanism was introduced into the feature fusion network to enhance feature extraction capabilities.Finally,EIOU was used as positioning loss function to improve the wafer detection accuracy.The experimental results show that the detection algorithm average accuracy is 99.3%,the parameter number is 4.637×10^(6),which achieves an ideal balance between the detection performance and the lightweight of the algorithm.

关 键 词：晶粒定位 目标检测 注意力机制 轻量级网络 损失函数 

分 类 号：TP39[自动化与计算机技术—计算机应用技术]