基于深度学习的包装缺陷快速检测方法

Rapid Packaging Defect Detection Method Based on Deep Learning

包装缺陷在线检测对提升产品质量具有重要意义。针对工业生产中人工目检效率低、漏检率高,以及基于显式特征提取的缺陷检测方法通用性差、特征提取复杂、缺陷区域占比小等问题,本文提出了基于迁移学习的改进MobilenetV2轻量化网络的包装缺陷快速检测方法,并利用某工厂糖果产线包装质检环节存在的4种表面缺陷对其进行测试。结果表明,该方法具有检测速度快、缺陷正确检测率高等优点,单个产品处理时间为0.053s,验证集上缺陷识别率为98.333%,在线测试缺陷品召回率为96.596%,在检测精度较高的同时能满足高速产线的实时性需求,最高支持2 m/s的传送带运行速度。

Online packaging defect detection is of great significance to improve product quality.Aiming al the problems of low detection efficiency,high missing detection rate of manual visual inspeetion and poor generality,complex feature extraction,ifcult application in complex backgrounds with small defeet areas of explicit feature extraction method in industrial production,a rapid detection method for packaging defects based on the transfer learming of improved MobileNetV2 lightweight network is proposed.Four surface defects in a candy production line are used to test the proposed method.The results show that the method has the characteristics of fast detection and high defect reeall rate.The processing time of a single product is 0.053s.the defect recognition rale on the verifcation set is as high as 98.33%,and the online test defects recall rate is 96.59%,The detection accuracy is relatively high.It can meet the real-time requirements of high-speed production lines at the same time,and support conveyor speed up to2 m/s.