基于机器视觉的聚氯乙烯管材表面缺陷检测

Surface Defect Detection of Polyvinyl Chloride Pipes Based on Machine Vision

针对人工肉眼检测聚氯乙烯(PVC)管材表面缺陷效果差、效率低下等问题,设计了一种基于机器视觉的PVC管材表面缺陷检测算法,并将其用于工业生产。该算法主要包含图像预处理和缺陷检测两部分,图像预处理包括边缘遍历、条纹检测和Gamma变换等步骤;缺陷检测主要包括水平与垂直投影、快速区域生长法连通域标记和分块处理等步骤。该算法对Gamma变换以及区域生长法作加速处理,同时能够最大限度地检测出PVC管材表面缺陷并避免误检。实验及工厂实地检测结果表明,该算法检测准确率为97.6%,实时检测速度超过60m/min,缺陷最小检测面积为0.05mm^2,而且管材运行中单边抖动不超过5mm时无误报警现象发生,管材在运行速度为45m/min时漏检率为0,因而能满足实际生产需要。

This study proposes a machine vision-based surface defect detection algorithm to enhance the effect and efficiency of detecting surface defects in polyvinyl chloride (PVC) pipes for industrial production. The algorithm performs image preprocessing and defect detection. Image preprocessing includes steps such as edge traversal, fringe detection, and Gamma transformation. Defect detection mainly includes horizontal and vertical projection, fast region growing for connected region marking, and block processing. The proposed algorithm accelerates the Gamma transformation and region growing, and it can also be used to optimally detect surface defects in PVC pipes, while avoiding false detection. Results of tests and actual factory inspections suggest that the proposed algorithm achieves a detection accuracy of 97. 6%, with a real-time detection speed of > 60 m/min, and a minimum defect detection area of 0.05 mm^2. Moreover, a unilateral jitter of <5mm does not cause any false alarms and the missed detection rate is 0 when the pipe runs at a speed of 45 m/min, which meets actual production needs.