差分拉曼光谱结合人工神经网络对药品塑料包装瓶的分类研究  

Classification of Plastic Pharmaceutical Packaging Bottles Based on Differential Raman Spectroscopy and ANN

作　　者：李锦 姜红[1] 杨俊 章欣 LI Jin;JIANG Hong;YANG Jun;ZHANG Xin(Detective College,People's Public Security University of China,Beijing 100038,China;Academy of Marxism,People's Public Security University of China,Beijing 100038,China;Nanjing Jianzhi Instrument and Equipment Co.,Ltd.,Nanjing 210049,China)

机构地区：[1]中国人民公安大学侦查学院,北京100038 [2]中国人民公安大学马克思主义学院,北京100038 [3]南京简智仪器设备有限公司,江苏南京210049

出　　处：《塑料工业》2022年第8期101-107,共7页China Plastics Industry

基　　金：中国人民公安大学2021年度基科费重点项目(2021JKF212);南京简智仪器设备有限公司合作项目(20191218)。

摘　　要：建立一个无损检验药品塑料包装瓶并对其进行分类的模型。利用差分拉曼光谱技术对47个样品进行检测,首先在原始数据的基础上进行差分拉曼光谱分析并进行人工分类,再运用Fisher判别法(FDA)和主成分分析法(PCA)对数据进行处理,结合人工神经网络算法(ANN-MLP/RBF)构建分类模型。在多层神经网络(MLP)模型中,使用原始数据、FDA处理后的数据、PCA降维后的数据对样本分类的正确率分别为87.23%、93.62%、97.87%,MLP模型下对样本分类的整体准确率为93%;在径向基神经网络(RBF)模型下,使用原始数据、FDA处理后的数据、PCA降维后的数据对样本分类的正确率分别为87.23%、93.62%、95.74%,RBF模型下对样本分类的整体准确率为92%。在研究相同条件下对药品塑料包装瓶进行分类时,采用PCA+MLP模型为最佳方案。A model for non-destructive testing and sorting of plastic pharmaceutical bottles was developed.47 samples were detected by differential Raman spectroscopy.Firstly, the differential Raman spectroscopy was analyzed and classified manually based on the original data.Then Fisher discriminant analysis(FDA)and principal component analysis(PCA)were used to process the data, and a classification model combined with artificial neural network algorithm(ANN-MLP/RBF)was established.In the multilayer neural network(MLP)model, the correct rates of sample classification using the original data, FDA-processed data, and PCA dimensionality reduction data are 87.23%,93.62%,and 97.87%,respectively.The overall accuracy of the sample classification under the MLP model is 93%.Under the radial basis function neural network(RBF)model, the correct rates of sample classification using the original data, FDA-processed data, and PCA dimensionality reduction data are 87.23%,93.62%,and 95.74%,respectively.The overall sample classification under the RBF model accuracy rate is 92%.When classifying pharmaceutical plastic packaging bottles under the same conditions as the study, the PCA+MLP model is the best solution.

关 键 词：差分拉曼光谱 人工神经网络 药品塑料包装瓶 无损检验 

分 类 号：TQ317.2[化学工程—高聚物工业]