基于卷积神经网络对纸质药品包装材料的红外光谱分析  

Analysis of Infrared Spectra of Paper Pharmaceutical Packaging Materials Based on Convolutional Neural Networks

作　　者：施磊 胡晓光[1] 姜红 莫修浩 Shi Lei;Hu Xiaoguang;Jiang Hong;Mo Xiuhao(School of Investigation,People's Public Security University of China,Beijing 100038,China;Center of Forensic Science Beijing Hui Zheng Zhuo Yue Technology Co.,Ltd,Beijing 102446,China;JINSP Company Limited,Beijing 100084,China)

机构地区：[1]中国人民公安大学侦查学院,北京100038 [2]北京汇正卓越科技有限公司司法鉴定中心,北京102446 [3]北京鉴知技术有限公司,北京100084

出　　处：《实验与分析》2025年第1期1-7,共7页LABOR PRAXIS

基　　金：中国人民公安大学刑事科学技术双一流创新研究专项(2023SYL06)。

摘　　要：为构建一种基于红外光谱技术的纸质药品包装材料分类模型,利用红外光谱技术对150种纸质药品包装材料样本进行检验,根据化学填料红外吸收峰的不同将样本人工分为三大类,根据分类结果筛选出6个特定波段,并采用方差归一化进行数据预处理,后构建一个基于残差和注意力改进的一维卷积神经网络(1D-CNN)模型。全部样本根据筛选波段划分为7个类别,将数据集按照3:7划分为训练集和测试集。基于注意力机制改进的一维卷积神经网络模型在所有波段的整体准确率表现良好,其中最突出的为All-select组,准确率达到了98.10%。基于红外光谱技术结合注意力机制改进的一维卷积神经网络(ATTN-1D-CNN)模型能够对纸质药品包装材料实现高效分类预测判别。To construct a classification model for paper-based pharmaceutical packaging materials using infrared spectroscopy technology,this paper employs infrared spectroscopy to examine 150 samples of paper-based pharmaceutical packaging materials.The samples are classified into three major categories based on the differences in infrared absorption peaks of chemical fillers.Based on the classification results,six specific wavelengths are selected,and variance normalization is applied for data preprocessing.Subsequently,a one-dimensional convolutional neural network(1D-CNN)model enhanced by residuals and attention mechanisms is developed.All samples are divided into seven categories according to the selected wavelengths and the dataset is split into training and testing sets at a ratio of 3:7.The test results indicate that the model exhibits good overall accuracy across all wavelengths,with the All-select group achieving the most notable accuracy rate of 98.10%.This demonstrates that the attention mechanism-enhanced one-dimensional convolutional neural network(ATTN-1D-CNN)model,combined with infrared spectroscopy technology,is capable of efficiently classifying and predicting paper-based pharmaceutical packaging materials.

关 键 词：药品纸质包装材料 红外光谱 注意力机制 一维卷积神经网络 

分 类 号：O657.37[理学—分析化学]