基于协同降维策略的电子鼻技术应用研究

Application Research of Electronic Nose Technology Based on Collaborative Dimensionality Reduction Strategy

本文提出了一种协同降维策略来优化特征维度进而提升电子鼻分类精度,该协同降维策略结合了无监督和有监督降维的优点实现原始特征的有效降维,并利用该策略实现不同品牌白酒的智能辨识。首先,提取电子鼻检测数据的最大值、稳态均值、积分值以及小波能量值作为特征值。其次,将无监督降维方式的核熵成分分析(KECA)引入对融合特征进行初步降维,再利用有监督降维方式的线性判别分析(LDA)进行再次降维得到最终的综合特征。最后,基于支持向量机(SVM)、概率神经网络(PNN)、随机森林(RF)对综合特征进行分类识别。结果表明,KECA-LDA-SVM获得了最高的分类性能达96%,说明该协同降维策略可以有效提升电子鼻的检测性能。

A collaborative dimensionality reduction strategy was proposed to optimize feature dimensions and improve the classification accuracy of electronic nose(e-nose).The collaborative dimensionality reduction strategy combined the advantages of the unsupervised and supervised dimensionality reduction.Firstly,the maximum value,steady-state average value,integral value and wavelet energy value of the e-nose detection data were extracted.Secondly,the kernel entropy component analysis(KECA)was introduced to perform the preliminary dimensionality reduction on the fusion feature,and the linear discriminant analysis(LDA)was used to perform the dimensionality reduction again to obtain the final comprehensive feature.Finally,based on the Support Vector Machine(SVM),Probabilistic Neural Network(PNN),Random Forest(RF)to classify the comprehensive features.The results shown that the KECA-LDA-SVM achieved the highest classification performance of 96%,indicating that the collaborative dimensionality reduction strategy can effectively improve the detection performance of the e-nose.