Sensor Array Optimization of Electronic Nose for Detection of Bacteria in Wound Infection

In order to identify the bacteria in wound infection, an electronic nose system with a sensor array of 34 sensors was designed. Eight kinds of samples were detected, i.e., culture medium, Escherichia coli, Staphylococcus aureus, Pseudomonas aeruginosa and their mixture with different concentration. Using support vector machine as the classifier and without sensor array optimization, the recognition rate is up to 86.54%. To simplify the sensor array and improve the recognition rate for bacteria samples, Wilks' lambda statistic (Wilks' Λ-statistic), Mahalanobis distance, principal component analysis (PCA), linear discriminant analysis (LDA), and genetic algorithm are used to optimize the sensor array. It is shown that the sensor array optimization may be realized efficiently by these methods except PCA. After sensor array optimization by Wilks' Λ-statistic and LDA, both of their recognition rates are the highest and up to 96.15%, while the numbers of sensors in optimized sensor arrays are 22 and 20, respectively. Under the limitation of ten sensors, the recognition rate optimized by Wilks' Λ-statistic and LDA may still reach 95.19%.