Feature Extraction And Pattern Classification Methods Study Based On Electronic Nose

Electronic nose as a mimic biological olfactory system of intelligent devices can be reliably and quickly realize the simple or complex odor discrimination. Compared with the traditional gas chromatography and other expensive gas analysis equipment, it is simple, reliable results, and for field testing, and therefore widely used in food, agriculture, health care, environmental monitoring and other industries. Feature extraction and pattern recognition are two key parts in an electronic nose system, which feature extraction significantly affect the reliability of the classification model and the accuracy of identification of unknown samples, pattern recognition is appropriate retreatment after feature extraction of information to obtain accurate information on gas composition and concentration. So advanced feature extraction and pattern recognition method and its application to electronic nose system has a high theoretical significance and practical value.This paper discusses the existing feature extraction and pattern recognition methods based on electronic nose system, study of the traditional principal component analysis, further study kernel principal component analysis on this basis, and then used these two methods to process two types of electronic nose data. In pattern recognition, it focuses on the support vector machine theory based on statistical learning theory, and use the method for qualitative identification of thirty different concentrations of CO, CH4, H2 gas composition, and quantitative analysis of different concentrations of CH4 gas samples; for the support vector machine kernel parameter selection error, obtained the optimal combination of parameters, introducing PSO to optimize the parameters of support vector machines. Finally, it used the algorithm based on the study of the kernel principal component analysis and support vector machines to process and discuss the high-dimensional data of electronic nose.Through the above studies suggest that:kernel principal component analysis of nonlinear problems or in dealing with the sample case of high dimension with good results, to extract more useful information; support vector machines compared to the structure of the complex neural network for classification and recognition process simple and fast, accurate and has a good classification and generalization ability, but also in quantitative analysis effectively reduces the relative error, and by particle swarm optimization algorithm can optimize the parameters of support vector machines further improve the prediction accuracy of the electronic nose; kernel principle component analysis and support vector machines can be effectively applied to electronic nose pattern recognition unit, for handling non-liear and high-dimension having a good effect, proving the feasibility of this method.