Study On Sensor Array Construction And Optimization Of Electronic Nose For Detection Of Bacteria In Wound Infection

For the detection of three target bacteria(i.e.Escherichia coli,Staphylococcus aureus and Pseudomonas aeruginosa)that are the most common bacteria in wound infection,a sensor array of electronic nose(e-nose)has been developed,which is combined with a sampling unit and a control unit.In order to improve the recognition rate for the three target bacteria and simplify the sensor array,optimization of sensor array has been studied.The main contents of the dissertation are as follows:Based on typical volatile metabolites produced by the most common bacteria in wound infection,metal oxide semiconductor(MOS)gas sensors and electrochemical sensors have been chosen,which are highly commercialized and mature.The signals of temperature,humidity,air pressure and working voltage were obtained and used to enhance anti-interference ability by response compensation.In addition,the structure of hexagonal prism has been devised as the sensor chamber which is made of Teflon.Experiments of bacterial data collection have been carried out based on the e-nose system for wound infection detection.Eight kinds of samples were detected,and a total of 480 sample data was obtained.In order to quantitatively evaluate the classification ability of original gas sensor array to three target bacteria,support vector machine(SVM)was employed to classify the experimental data.The results showed that the recognition rate of testing set was up to 86.54%,which indicated that the sensor array could distinguish Escherichia coli,Staphylococcus aureus and Pseudomonas aeruginosa effectively.To improve the recognition rate for bacteria samples and simplify the sensor array,Wilks lambda statistic(Wilks ?-statistic),Mahalanobis distance(MD),principal component analysis(PCA),linear discriminant analysis(LDA)and genetic algorithm(GA)were used to optimize the sensor array with different ways(i.e.variable selection,data dimensionality reduction and sensor combination search).The recognition rate of testing set obtained by SVM and the number of sensors in the optimized sensor array were used as main indices to evaluate the optimized sensor arrays and corresponding sensor array optimization methods.Among the five methods used in this thesis,good results were obtained by the sensor array optimization methods based on Wilks ?-statistic and LDA.In the case of unfixed number of sensors,the best recognition rate obtained by LDA was up to 96.15%,while the number of corresponding optimized sensors was 20.With the limitation of 10 selected sensors,the best recognition rate obtained by sensor array optimization methods based on Wilks ?-statistic or LDA was up to 95.19%,while the latter performed better in all data set(i.e.training set,validation set and testing set).