基于FastICA和神经网络的电子鼻模式识别

气体传感器阵列是电子鼻系统的重要组成部分,传感器阵列的交叉敏特性严重影响电子鼻对气体识别的准确率.将快速独立分量分析算法和BP网络相结合用于电子鼻的模式识别可以有效地改善这一问题.并由一个5个传感器组成的电子鼻系统,对10组不同体积分数的3种气体测量得到的30组数据样本进行仿真.结果表明,用快速独立分量分析对数据作预处理,可以简化计算,减少数据之间的相关性,将预处理后的数据样本作为BP网络的输入,使网络结构简化,收敛速度快.利用该方法可以提高电子鼻识别气体的准确率.

Study on electronic-nose recognition method with fast independent component analysis and neural network

Gas sensor array was an important part of electronic nose.The recognition of electronic nose was affected badly by the cross sensitivity of gas sensor array.This problem can be improved while fast independent component analysis(FastICA) and back propagation neural network(BP NN) were used for pattern recognition of electronic nose.The various of volume fractions of three sorts of gases was analyzed by experiments using a sensor array of five sensors.The results showed that Fast-ICA can predigest the calculator consumedly and reduce the data correlation,the data processed was inputted to BP Network,the network was simplified and the convergence speed was enhanced greatly.Further the recognition was improved highly with FastICA and BP network to Electronic-nose system.