Study On Sensor Array Optimization Of Medical Electronic Nose For Wound Infection Detection

The construction and design of electronic nose system originates from biological olfactory system,in which gas sensor array,signal preprocessing and pattern recognition simulate olfactory receptor cells,olfactory bulb and cerebral cortex in biological olfactory system respectively.As the core component of electronic nose,the number of sensors in gas sensor array is often relatively large in the initial design and fabrication process.More gas sensors will have two adverse effects: first,the redundant information and the noise information caused by and environmental factors will increase,which will lead to the decline of the performance of electronic nose system;second,the design of circuit will be more complex,the cost of making electronic nose system will increase,and the possibility of introducing noise into complex circuit design will be higher.Therefore,research on sensor array optimization is indispensable in electronic nose technology.In order to improve the detection effect of the two sets of electronic nose system and simplify the number of sensors in the two system,methods based on factor analysis and Hilbert-Schmidt independent criterion were proposed to optimize sensor array.The classification results of BP neural network and support vector machine(SVM)were compared in this thesis,and SVM was chosen as the pattern recognition classification algorithm of two electronic nose systems designed by our laboratory.The recognition rate of test set on bacterial culture medium data set was 83.94%,and the recognition rate of test set on animal wound infection data set was 93.29%.In the two optimization methods of sensor array,weighted factor analysis method and non-weighted factor analysis method were proposed based on factor analysis.The weighted factor analysis method is to multiply the contribution rate of the selected common factor by the corresponding factor load matrix to generate a new factor load matrix,and then the sensors is ranked accordingly.Experimental results show that the optimization effect of weighted factor analysis was better.The recognition rate of test set on bacterial culture medium data set was 93.08%(compared with the unoptimized state,the number of sensors was reduced by 15),and 94.41%(the number of sensors was reduced by 3)on animal wound infection data set.The principle of optimization method based on Hilbert-Schmidt independent criterion is to select features according to the relevance between features and labels.In this thesis,the Hilbert-Schmidt independent criterion optimization method of linear kernel function and Gaussian kernel function was used.Experimental results show that the Hilbert-Schmidt independent criterion optimization method with Gaussian kernel function had better optimization effect.The recognition rate of bacterial culture data set was 93.65%(the number of sensors was reduced by 16),and that of the animal wound infection data set was 94.53%(the number of sensors was reduced by 13).Of the two sensor array optimization methods(factor analysis optimization and Hilbert-Schmidt independent criterion optimization),Hilbert-Schmidt independent criterion optimization method had better optimization effect.Compared with the linear discriminant analysis method,the weighted factor analysis optimization method and Hilbert-Schmidt independent criterion optimization method(Gaussian kernel function)were better.