计算机视觉结合深度学习技术快速鉴别八角粉掺伪  

The application of computer vision combining with deep leaning techniques for rapid discrimination of adulterated star anise powder

作　　者：陈劲星 CHEN Jinxing(Fujian CCIC-Fairreach Food Safety Testing Co.,Ltd.,Fuzhou,Fujian 350008,China)

机构地区：[1]福建中检华日食品安全检测有限公司,福建福州350008

出　　处：《食品与机械》2023年第12期42-47,69,共7页Food and Machinery

摘　　要：目的:设计一种基于计算机视觉技术结合深度学习模型的新方法检测八角粉的掺假情况。方法:采集不同掺假比例八角粉的原始图像,利用预处理和数据增强技术获得图像集合。随后构建SqueezeNet深度学习模型,并与支持向量机(support vector machine, SVM)、K-邻近学习(K-nearest neighbor learning, KNN)、随机森林(random forest, RF)、梯度提升树(gradient boosting tree, GBT)和多层感知器(multilayer perceptron, MLP)5种机器学习模型进行比较。结果:5种机器学习模型的最高准确度仅为66.37%,而SqueezeNet模型的准确度为99.42%。结论:深度学习分类模型性能相较于传统机器学习分类模型更为优越,识别效果良好且样品无需预处理。Objective:This study aims to design a novel approach,utilizing computer vision combining with deep learning,for rapid determination the adulteration in star anise powder.Methods:Collected the original images of star anise powder with varying adulteration ratios.Employing preprocessing and data enhancement techniques,an image dataset was curated.Subsequently,a SqueezeNet model was constructed and compared with five machine learning models,including Support Vector Machine(SVM),K-Nearest Neighbor Learning(KNN),Random Forest(RF),Gradient Boosting Tree(GBT),and Multilayer Perceptron(MLP).Results:The highest accuracy achieved by the five machine learning models was only 66.37%,while the accuracy of the SqueezeNet model was 99.42%.The results showed that SqueezeNet model was better than these machine learning models in identifying the adulteration in star anise powder.Conclusion:The proposed detection method based on computer vision combining with SqueezeNet model can effectively identify the adulteration in star anise powder.This method is easy to operate,and provides a novel technique for the rapid detection of food adulteration.

关 键 词：八角 掺伪鉴别 深度学习 视觉技术 SqueezeNet模型 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TP391.41[自动化与计算机技术—控制科学与工程] TS264[轻工技术与工程—发酵工程]