Comparative Research On SVM Optimization Method Based On Multi-class Image

SVM (Support Vector Machine SVM) has been widely applied to texture imagesegmentation. However SVM parameters appropriate or not will have an influence on thetraining and convergence of the data. Currently the mainly method of optimizing SVMparameter is swarm intelligence algorithm. But the generalization of swarm intelligenceoptimization algorithm is not good. To solve this problem, improving the existing intelligentalgorithms and mixing different swarm intelligence algorithms to create new hybrid intelligentalgorithm have become hot topics. This paper studies this two aspects, and improving PSOalgorithm, and mixing with other swarm intelligence algorithms, then PSO algorithm is usedto optimize parameters of support vector machine after the performance of the algorithm hasbeen improved. The optimized SVM model will eventually be applied to different textureimages segmentation.Firstly, this paper introduces the basic principles of support vector machine and the typeof current existing support vector machine. Then this paper analyzes the impact of modelparameters on the performance, improving traditional PSO algorithm inertia weight. Finally,the improved particle swarm optimization will be applied to optimize the support vectormachine parameters. To validate texture image segmentation, in this paper, two texture imageswith different sources, namely electron microscopy images and ultrasound image will bedivided. Experimental results show that the segmentation result of the improved PSOalgorithm by dividing the texture image is better than segmentation results by using geneticalgorithms and grid method obtained.Secondly, in order to improve the robustness of the optimization algorithm, the paperimproves niche and cross-selection operator PSO. First mutation mechanism is introduced intothe algorithm. The particles which have bad fitness value will be executed crossover mutationselection operation. Then group competition mechanism is introduced into the algorithm andthere is competition for operation among different populations of particles. Finally parallelmixing the improved niche and cross-selection operator PSO and artificial fish swarmalgorithm.The mixed algorithm is used to optimize the parameters of support vector machine.Experimental results show that the hybrid algorithm with the improved niche andcross-selection operator PSO and artificial fish swarm algorithm which is used to optimizeSVM parameters. Compared to the experimental results of the niche and cross-selectionoperator PSO, improved PSO algorithm, artificial fish swarm algorithm. This result of twokinds of texture image segmentation with the method is very ideal, segmentation modelstability relative increase.In order to better illustrate the advantages of optimization of SVM parameters of themethod in this page, this paper discussed four kinds of traditional algorithm parametersoptimization: Artificial bee colony algorithm, Ant colony optimization algorithm, Differential evolution algorithm, Simulated annealing. The four algorithms and the niche andcross-selection operator PSO are mixed to form a hybrid algorithm. These eight algorithms areused to optimize the SVM parameters, The optimized SVM classifier will be used for dividingtwo kinds of texture image. Finally, texture image will be reconstruction according to the label,The experimental results show that the this method can get better SVM parameters, and isgood for texture image segmentation, Not only makes the segmentation results are better thanthe other algorithms, but also have a great advantage of the robustness of the algorithm.