基于无监督深度学习的卷烟包装外观缺陷检测

Appearance Defect Detection of Cigarette Packaging Boased on Unsupervised Deep Learning

针对现有卷烟包装外观缺陷检测方法所存在的稳定性差、适用性不强等问题,提出一种基于无监督深度学习的卷烟包装外观缺陷检测方法。该方法基于轻量化改进后的PatchCore算法,只训练正常样本即可实现对卷烟包装外观缺陷的检测。通过实验最终对比原始PatchCore,检测精度从96.84%提升到100%,检测时间从92.10 ms缩短到69.25 ms,实现了检测精度及速度双高的卷烟包装外观缺陷检测,具有一定的实用意义。

To solve the problems of existing cigarette packaging appearance defect detection method such as poor stability and weak applicability,a cigarette packaging appearance defect detection method based on unsupervised deep learning is proposed.The method,based on the lightweight and improved PatchCore algorithm,only trains normal samples to detect the appearance defects of cigarette packaging.The results of final experimental comparison with the original PatchCore indicate that the detection accuracy increases from 96.84%to 100%and the detection rate from 92.10 ms to 69.25 ms,which shows that the proposed method can realize the detection of cigarette package appearance defects with high detection accuracy and high speed,and has certain practical significance.