纸基比色阵列传感器结合机器学习识别不同年份黄酒  

Paper-Based Colorimetric Array Sensors Combined with Machine Learning Identify Different Vintages of Huangjiu

作　　者：梁世有 赵世茹 王珍珍 戴静 沙如意 毛建卫 LIANG Shiyou;ZHAO Shiru;WANG Zhenzhen;DAI Jing;SHA Ruyi;MAO Jianwei(School of Biological and Chemical Engineering,Zhejiang University of Science and Technology,Hangzhou,Zhejiang 310023)

机构地区：[1]浙江科技大学生物与化学工程学院,浙江杭州310023

出　　处：《核农学报》2025年第4期793-799,共7页Journal of Nuclear Agricultural Sciences

基　　金：省属高校基本科研业务费专项资金(2023JLYB007)。

摘　　要：为解决传统黄酒年份识别方法耗时长、成本高的问题,本研究首次运用纸基比色阵列传感器技术,借助线性判别分析(LDA)处理黄酒样本特征数据,对不同年份黄酒进行基础分类;应用主成分分析(PCA)对黄酒样本的特征数据进行处理和降维;采用K近邻(KNN)和梯度提升树(GBT)两种机器学习算法建立分类模型。结果表明,当GBT模型参数主成分数为3、决策树数量为150、每颗树的最大深度为3时,模型的测试集分类准确率高达97.5%,五折交叉验证准确率为95.7%;KNN模型参数主成分数为7、K值为3时,模型的测试集分类准确率为93.75%,五折交叉验证准确率为90%;利用GBT模型进行随机验证,识别准确率为95%,表明识别效果优良。本研究可为不同类型黄酒年份检测提供新的解决思路与方案。In order to solve the problems of long time and high cost in traditional Huangjiu vintage identification methods,this study first used paper-based colorimetric array sensor technology to process the characteristic data of Huangjiu samples with linear discriminant analysis(LDA)for basic classification of different years of Huangjiu.Principal component analysis(PCA)was applied to process and reduce the dimensionality of the characteristic data of Huangjiu samples.Two machine learning algorithms,K-nearest neighbor(KNN)and gradient boosting tree(GBT),were used to establish classification models.The results showed that when the GBT model parameters had a principal component number of 3,a decision tree number of 150,and a maximum depth of 3 per tree,the model’s test set classification accuracy reached 97.5%,and the five-fold cross-validation accuracy was 95.7%.When the KNN model parameters had a principal component number of 7 and a K value of 3,the model’s test set classification accuracy was 93.75%,and the five-fold crossvalidation accuracy was 90%.Random validation using the GBT model showed an identification accuracy of 95%,indicating excellent recognition performance.This study can provide new solutions for different types of Huangjiu vintage detection.

关 键 词：黄酒 纸基比色阵列传感器 分类识别 机器学习 

分 类 号：TS262.4[轻工技术与工程—发酵工程] O657.3[轻工技术与工程—食品科学与工程] TP181[理学—分析化学]