基于改进Yolov5网络的经真空干燥皮革图像缺陷检测  

Defect Detection of Vacuum Dried Leather Images Based on Improved Yolov5 Network

作　　者：陈宏[1] CHEN Hong(Xi'an Eurasia University,Xi'an 710065,China)

机构地区：[1]西安欧亚学院,陕西西安710065

出　　处：《皮革与化工》2025年第2期12-17,共6页Leather And Chemicals

摘　　要：为更好调节皮革真空干燥机温度控制效果,提出一种基于改进Yolov5网络的经皮革真空干燥机干燥的皮革图像缺陷检测方法。其中,首先对Yolov5网络的主干网络和损失函数进行改进,然后采用改进的Yolov5网络对经皮革真空干燥机干燥的皮革图像进行缺陷检测;最后将该网络用于经皮革真空干燥机干燥的皮革裂缝图像等的缺陷检测中。结果表明,改进后的Yolov5网络的精确率、召回率和m A P分别取值为96.47%、95.19%和97.32%,相较于改进前,分别提升14.15%、28.02%和19.63%;对比于传统的卷积神经网络和改进双边滤波检测算法,本网络的精确率、召回率和m A P明显更高。综合分析可知,采用本网络可实现对皮革真空干燥机干燥后的皮革进行图像检测,从而为后续的干燥机温度控制提供参考。In order to better regulate the temperature control effect of leather vacuum dryer,a image defect detection method based on improved Yolov5 network is proposed for leather dried by leather vacuum dryer.Firstly,the backbone network and loss function of the Yolov5 network are improved,and then the improved Yolov5 network is used to detect defects in images of leather dried by a leather vacuum dryer;Finally,the network is applied to defect detection of cracked images of leather dried by a leather vacuum dryer.The results showed that the accuracy,recall,and mAP values of the improved Yolov5 network were 96.47%,95.19%,and 97.32%,respectively.Compared with before the improvement,they increased by 14.15%,28.02%,and 19.63%,respectively;Compared to traditional convolutional neural networks and improved bilateral filtering detection algorithms,this network has significantly higher accuracy,recall,and mAP.Based on comprehensive analysis,it can be concluded that using this network can detect the leather images after drying in a leather vacuum dryer,providing reference for subsequent temperature control of the dryer.

关 键 词：皮革真空干燥机 机器识别 Yolov5网络 缺陷检测 主干网络 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]