A Novel Type of Feature Extraction Algorithm Based on Multi-Directional Analysis for Visual Optical Electronic Nose

Abstract

According to the fact that the response images of a visual optical electronic nose (E-nose) have a huge amount of data, various frequency components and complex periodic and directional information, a novel type of visual optical E-nose (VOE-nose) feature extraction algorithm based on multi-directional analysis by directional filter bank (DFB) was proposed in this paper. Firstly, the gas sensing model of the VOE-nose was introduced, and the basic principle of DFB algorithm for feature extraction was described. Second, response images of NO2 in different wavebands were collected by the VOE-nose platform. Third, typical feature extraction and DFB feature extraction algorithms were used to extract features of response data, then the feature dimension reduction and pattern recognition algorithms were used to analyze the features. The mean classification accuracy is more than 95%, which verifies the superiority of the DFB feature extraction algorithm.