相似度分析结合人工神经网络鉴别苹果香精  

Identification of apple essences based on similarity analysis and artificial neural network

作　　者：王海燕[1,2] 袁雪琴 宋超 张正勇[1,2] 刘军[1,2] 沙敏 

机构地区：[1]南京财经大学管理科学与工程学院,南京210046 [2]江苏省质量安全工程研究院,南京210046

出　　处：《计算机工程与应用》2018年第4期141-147,共7页Computer Engineering and Applications

基　　金：质检公益性行业科研专项(No.201410173);国家重大科学仪器设备开发专项(No.2013YQ090703);国家自然科学基金面上项目(No.61373058);国家自然科学基金重点项目(No.71433006);国家自然科学基金(No.61602217)

摘　　要：采用相似度分析结合人工神经网络的方法鉴别苹果香精。香精经水溶液稀释处理后再经离子迁移谱仪分析,针对仪器波动性引起的谱图差异,应用相似度分析的方法判别离群谱图,筛选有效谱图,建立了香精样品的离子迁移谱指纹图谱库。通过主成分分析提取香精指纹图谱的特征向量输入误差反向传播神经网络,建立了几种不同网络结构的分类鉴别模型。实验结果表明,采用单层感知机,网络结构为10-13-5,激活函数为tansig,训练函数为trainbr的分类鉴别模型最佳,能避免过拟合,获得更加精确的识别效果,识别率达99.41%,且较去离群谱图前的识别率提高了3.82%。Similarity analysis and artificial neural network method are combined to identify apple essences. This paper uses Ion Mobility Spectrometry(IMS)equipment to analyze essence after diluted by aqueous solution, to be directed against the outliners came from volatility of IMS equipment test, it uses similarity analysis method to discriminate outliners, screen valid samples and establish IMS fingerprint database of essence samples. Extract feature vectors of essence fingerprint through principle component analysis, input them to error back propagation neural network and establish several identification models of different network structures. Experimental results show that the model using single layer perception, of10-13-5 network structure, tansig activation function, trainbr training function is the best classification model. It can avoid overfitting and achieve more precise recognition effect. The recognition rate can reach 99.41%, 3.82% is improved compared to when outliners are not removed.

关 键 词：相似度分析 人工神经网络 离子迁移谱 指纹图谱 鉴别模型 苹果香精 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]