Research And Application On SVM Kernel Parameters Optimization

SVM(Support Vector Machine, SVM) was proposed in the early1990s by Vapnik, it’s based on the statistical learning theory under the framework of a new pattern recognition method. SVM is good at solving pattern recognition classification problems such as small sample, nonlinear, and can be applied to the function fitting and other machine learning. The SVM classification performance mainly depends on the choice of kernel function and the releated parameters. Improper parameters will lead to a classification performance even unable to classify the sample.The current research on SVM kernel function and its parameters are mor and gradually mature, but has not yet formed a unified method of kernel function parameter selection. In most cases only can rely on experience or contrast experiment to set parameters, which is a theory in the practical application of SVM. In view of this, the research content of this paper is the ways of the selection of kernel functions for SVM parameters.This article mainly discussed the following aspects of content:First introduced the basic theory of the SVM system comprehensively, including the statistical learning theory, the theory of VC dimension and structural risk minimization principle, etc.Secondly, discuss and analysis of several common kernel functions for SVM parameters selection method, including the grid search method, as well as the merits and demerits of the current popular particle swarm search method, at the same time, based on the above analysis puts forward a new method of kernel functions for SVM parameters selection-improved glowworm bionic algorithm. Through experiments and analysis in this paper, the proposed algorithm in accuracy and operation time compared with other algorithms have the advantage and feasibility.Finally I put the optimized SVM classification used in space handwriting recognition. Through the acquisition of acceleration sensor data preprocessing and feature extraction, and finally using the optimizated classification. Experimental results show that the proposed algorithm presented in this paper to optimize the SVM classification equipment have good classification performance.