Detection of Hazardous Gas Mixtures in the Smart Kitchen Using an Electronic Nose with Support Vector Machine

Abstract

The detection of hazardous gases are essential to protect human health and safety. Nowadays, there is a great demand for the detection of multiple hazardous gases. In this study, a miniaturized electronic nose with SVM recognition models was used for the detection of carbon monoxide, methane, formaldehyde as well as their mixtures. The sensor array consisted of 6 commercial MOS sensors which were cross-sensitive to three kinds of hazardous gases. The SVM models were trained based on the features extracted by two methods in order to recognize the concentration levels of three hazardous gases. The 5-fold cross-validation was used to evaluate and compare the accuracies of different models for all target gases. The results indicated that the wavelet time scattering can extract features more effectively compared with the classic feature extraction method. The models based on the features gained by wavelet time scattering showed the accuracies of 98.73% for CO, 100% for CH4 and 97.46% for CH2O. This study provides a practical recognition method and detection platform for multi-gas sensing applications.