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

Abstract: 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, of 10-13-5 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.

Key words: similarity analysis, artificial neural network, Ion Mobility Spectrometer（IMS）, fingerprint, identification model, apple essence