An entire feature extraction method of metal oxide gas sensors

A phase space entire (PSE) feature extraction method for metal oxide gas sensors was proposed in this paper. This method extracts six features from the response curve in phase space. The difference from other feature extraction methods such as extracting phase space integral (PSI), which also use phase space, is that the sensor response curves could be completely described and well reconstructed with the six features by a sensory response feedback process. The proposed method PSE was compared with the PSI extracting method and a common signal maximum (SM) extracting method in a dataset. The dataset was consisted of 264 measurements of 11 volatile organic compounds (VOCs) of four concentrations by six TGS gas sensors. The six features were first extracted from each response curve of all samples, and then all response curves were reconstructed by the six features. The mean sensory response feedback error of the reconstructed response curves from the original signal response curves was 5.4%, which indicated that PSE could extract the entire signal of response curves. Meanwhile, Fisher discriminant analysis (FDA) was used in the comparison of the three feature extraction methods. The correct recognition rates of FDA with SM, PSI and PSE were 90.9%, 81.1% and 100%, respectively. The results show that PSE is an efficient entire feature extraction method.