基于轻量化和注意力机制改进YOLOv5网络的模具缺陷识别  

Improvement of YOLOv5 network for mold defect recognition based on lightweight and attention mechanism

作　　者：杨飞 Yang Fei(School of Information Engineering,Zhengzhou Urban Construction Vocational College,Zhengzhou 450052,Henan,China;School of Softuare,Xi'an University of Electronic Science and Technology,Xi'an 710126,Shaanxi,China)

机构地区：[1]郑州城建职业学院信息工程学院,河南郑州450052 [2]西安电子科技大学软件学院,陕西西安710126

出　　处：《模具技术》2025年第1期88-96,共9页Die and Mould Technology

摘　　要：为提高模具缺陷识别的精度和速度,提出一种基于改进YOLOv5网络的识别方法。该方法采用PP-LCNet轻量化网络作为主干网络,F-CIoU函数作为损失函数,并引入CBAM注意力机制,对YOLOv5网络进行改进。然后采用改进YOLOv5网络对不同类型的模具缺陷进行识别。结果表明,该方法具有较高的模具缺陷识别精度和速度,对凹槽和污点模具缺陷识别的平均查全率、查准率、平均精度均值、F1值分别达到95.34%,95.03%,93.11%,95.68%,平均检测速度达到35帧/s,满足模具缺陷识别的精度和实时识别需求。To improve the accuracy and speed of mold defect recognition,a recognition method based on an improved YOLOv5 network is proposed.We use PP-LCNet lightweight network as the backbone network,F-CIoU function as the loss function,and introduce CBAM attention mechanism to improve YOLOv5 network.Then,an improved YOLOv5 network was used to identify different types of mold defects.The results show that this method has high accuracy and speed in identifying mold defects.The average recall,precision,average accuracy,and F1 value for identifying groove and stain mold defects reach 95.34%,95.03%,93.11%,and 95.68%,respectively.The average detection speed reaches 35 frames per second,meeting the accuracy and real-time recognition requirements of mold defect recognition.This provides a reference for improving the accuracy and speed of identifying different mold defects.

关 键 词：模具缺陷 缺陷识别 YOLOv5网络 CBAM注意力机制 PP-LCNet轻量化网络 

分 类 号：TP389[自动化与计算机技术—计算机系统结构]