A Bioinspired Neural Network for Data Processing in an Electronic Nose

Abstract

A novel bioinspired neural network is proposed as a replacement for traditional data processing methods in the electronic nose (e-nose) that we designed. This neural network mimics the main structures of the mammalian olfactory system. It contains olfactory sensing neurons, mitral cells, and granule cells. The proposed method can directly use the raw data collected from sensors in the e-nose without any signal preprocessing, feature selection, or reduction. The output neurons of the network can change the sensors’ responses into two new time series, of which we only use the variances to perform classification. This significantly simplifies the data processing procedure in e-noses. In order to test the performance of the proposed bioinspired neural network in the e-nose, two sampling methods, three classification methods, and seven kinds of Chinese liquors were employed. The highest classification rates of the proposed method and traditional method are 100% and 93%, respectively.