基于激光散斑的食品包装封口异常检测

Anomaly Detection of Food-Packaging Seals Based on Laser Speckle

摘要: 为克服当前食品包装袋质量检测手段的局限性，丰富食品企业的包装质量检测方法，搭建了激光散斑图像采集系统，采用暗场照射方式获取封口的激光散斑图像。针对采集的图像，使用阈值分割、高斯滤波和对比度限制的自适应直方图均衡化进行预处理；随后对缩放后的图像采用完全主成分分析(Complete Principal Component Analysis, CPCA)进行特征降维，并利用单类支持向量机(One Class Support Vector Machine, OCSVM)对降维后的特征向量进行分类识别。实验结果表明，采用高斯径向基核函数(Radial Basis Function, RBF)时，检测效果最优，灵敏度达到94.59%，特异性达到97.30%，精确率达到95.95%，接收者操作特征曲线下面积(Area under Curve, AUC)达到98.24%。该研究证实，激光散斑结合机器视觉技术可凸显封口纹理特征，有效提升食品包装封口的成像质量，具有实际应用价值。

Abstract: To overcome the limitations of current quality inspection methods for food packaging bags and enrich the quality testing approaches for food enterprises, a laser speckle image acquisition system was constructed. Dark-field illumination was employed to obtain laser speckle images of the seals. For the acquired images, preprocessing was performed using threshold segmentation, Gaussian filtering, and Contrast Limited Adaptive Histogram Equalization. Subsequently, the scaled images underwent feature dimensionality reduction via Complete Principal Component Analysis (CPCA), and the reduced feature vectors were classified and recognized using a One-Class Support Vector Machine (OCSVM). Experimental results indicated that the optimal detection performance was achieved when employing the Gaussian Radial Basis Function (RBF) kernel, with a sensitivity of 94.59%, specificity of 97.30%, precision of 95.95%, and an Area under the Receiver Operating Characteristic Curve (AUC) of 98.24%. This study demonstrates that laser speckle technology combined with machine vision can effectively highlight the texture features of seals and significantly enhance the imaging quality for food packaging seal inspection, confirming its practical application value.