Research Of LS-WSVM Model Based On Binary Tree In Early Fire Classification

Fire has been one of the common destructive and most influencing disasters in our country, when fire happens, people will be hurt and treasure will be loss hugely. So it is very significant to study fire fore-warning. In order to discover fire earlier and control fire spreading, we use electronic nose combining with gas sensors to collect the fire information when fire is in smoldering situation. And then we recognize the characteristic information and achieve the goal of distinguishing the early fire classes, we estimate the cause of fire and it can help us to know how to put out the fire efficiently and quickly. This paper's study is based on summarizing early fire fore-warning home and abroad. Least squares wavelet support vector machine multi-class classification model based on binary tree was proposed to avoid the shortage of algorithm before. We use principal component analysis method to extract the feature of early fire information which was collected by electronic nose sensors array, at last the feature information was recognized by multi-class classification model, the distinguishing and classification of fire is come true, and then we achieve our purpose of fire early fore-warning.This paper first introduces the theory of support vector machine and how to structure least squares wavelet support vector machine. Second, we study the method of nonlinear mapping and combine it with binary tree structure and least squares wavelet support vector machine, then least squares wavelet support vector machine model based on binary tree was proposed. To consider the characteristics of early fire information, principal component analysis method was proposed to extract the characteristics of early fire information. Finally, least squares wavelet support vector machine model based on binary tree was used to the experiment of early fire classification, and compared with BP neural network, K-Nearest Neighbor algorithm and decision tree algorithm, the results show the model has higher recognition rate in the early fire classification. Support vector machine using wavelet kernel function has lower training time, classification time and higher recognition rate compared with support vector machine using radial basis kernel function. The classification model based on balanced binary tree has higher training speed and classification speed. The early fire classification experiments show least squares wavelet support vector machine model based on binary tree has better recognition effect and faster classification speed; it is fitter for the application of early fire classification.