基于电子鼻的深度卷积神经网络茯苓产地分类方法  

DCNN for poria cocos origin classification method based on E-nose

作　　者：徐昊 章检明 王中鹏[1] 张丽娟 迟梁 何成 XU Hao;ZHANG Jianming;WANG Zhongpeng;ZHANG Lijuan;CHI Liang;HE Cheng(School of Information and Electronic Engineering,Zhejiang University of Science and Technology,Hangzhou 310000,China;Zhejiang Academy of Agricultural Sciences,Institute of Food Science,Hangzhou 310000,China)

机构地区：[1]浙江科技学院信息与电子工程学院,浙江杭州310000 [2]浙江省农业科学院食品科学研究所,浙江杭州310000

出　　处：《传感器与微系统》2023年第12期138-141,共4页Transducer and Microsystem Technologies

基　　金：浙江省“三农九方”科技协作计划资助项目(2023SNJF007)。

摘　　要：本文采用自研电子鼻系统,实现6个产地的茯苓气味样本的检测,根据采集得到的样本数据集,提出并优化了适用于茯苓产地分类的卷积神经网络—长短期记忆网络(CNN-LSTM)模型,同时与支持向量机(SVM)模型、CNN模型、LSTM模型进行对比,识别率提高6%以上。在实际样本检测中,适用于茯苓产地分类的CNN-LSTM模型识别准确率为81.9%,优化后的CNN-LSTM模型识别准确率达到了88.9%,且优化后的神经网络能够更快、更好地从电子鼻数据中提取特征。A self-developed electronic nose(E-nose)system is used to conduct odor sample detection of poria cocos from six different origins.Based on the collected sample dataset,a convolutional neural network long short-term memory(CNN-LSTM)model for classifying the origin of poria cocos is proposed and optimized.Compared with SVM,CNN,and LSTM models,the result shows that recognition rate is increased by over 6%.In practical sample detection,the CNN-LSTM model suitable for poria cocos origin classification achieves a recognition accuracy of 81.9%,while the optimized CNN-LSTM model reaches 88.9%.The optimized network can extract features more quickly and effectively from E-nose data.

关 键 词：电子鼻 深度学习 中药产地区分 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]