A Novel Feature Extraction Technique of Electronic Nose for Detecting of Wound Infection Based on Phase Space

Abstract

Rapid and timely monitoring of traumatic inflammation is conducive to doctors? diagnosis and treatment. It has been proved that electronic nose (E-nose) is an effective way to predict the bacterial classes of wound infection by smelling the odor produced by the metabolites, and it has also been found thatthe classification accuracy of E-nose is very different when different feature is extracted and put into the classifier. The gas sensor array of E-nose can be seen as a dynamic system whose response temporally evolves following the concentration of the odors. As the central concept in the analysis of dynamic systems, phase space is the first time to be employed by us to construct the feature matrix of wound infection data in this paper. Dynamic moments, the functions of time delay in phase space, is used as the feature of wound infection. The odors of four different classes of wound (wound uninfected, and infected withP. aeruginosa, E. coliandS. aureus) are used as the original response of E-nose.Experimental results prove that the classification accuracy of test data set is 96.43% when R2 is used as the feature, which is much better than M2P, M3P (other two dynamic moments), maximum value of the steady-state response and maximum value of the first-order derivative (two traditional feature of E-nose).