Optimization Parameters Based On SVM Classification Of Remote Sensing Image

Genetic Algorithm and Particle Swarm algorithm are belong to the category of SwarmIntelligence algorithm, its basic idea is through the computer technology to imitate some of thebehavior of animals, is used to solve complicated programming that can solve the problems in the reallife.Its important characteristic is extremely complex and difficult to solve the problems in theplanning, as well as no need to specific function relation between derivation, but combined with theoptimal function completed parameter programming solver.Genetic Algorithms in theimplementation process which has some of characteristics such as easy to understand, easy toimprove, have been belong to machine learning. Other fields have been more widely used.SVM (Support Vector Machine, SVM) classification method solving the problems such as smallsample, nonlinear, local minimum value, has a unique advantage on the structural risk minimizationthat is good at avoiding the problem of learning, therefore has been widely used in many classificationresearch field.Reserach using the SVM classifier has carried on the classification of remote sensingimage, is realized in MATLAB about Genetic Algorithm and Particle Swarm Algorithm to optimizethe punishment and kernel parameter of SVM classification. Genetic Algorithm and Particle SwarmOptimization based on radial basis kernel function(RBF) of SVM classifier parameters optimizationThis paper has done research work is following:(1) In the MATLAB IDE, Research have been realizing the function of the SVM with LibSVMsoftware package. The package has carried on the detailed introduction, mainly on Cross Validation(Cross Validation) in the parameter Settings. As well as it was realizing the classification of the image,sample data from format conversion, model training. Image classification research has achieved in thedetailed introduction.(2) The implementation mechanism of Genetic Algorithm and Particle Swarm Optimizationalgorithm was analyzed in the MATLAB environment with LibSVM cross validation model asfitness function. Research realized the parameters optimization of Genetic Algorithm and ParticleSwarm Optimization, including the selection, exchange method and mutation in GAstages.Integrating with LibSVM classification tool, reserach archieved two kinds of reasonableparameter optimization algorithm for the study of remote sensing image classification of SVM in thestudy region. (3) The study region is in xishuangbanna GuanLei town and mong-la nature reserve, therebyformulated the classification system based on forest inventory. It has completed the parametersoptimization of GA and PSO on SVM classification of remote sensing image. The classificationresults were evaluated by fuzzy evidential accuracy. The overall accuracy of Genetic Algorithmoptimization is81.1044%. The overall accuracy of PSO optimization is85.0969%. Compared twokinds of results, population fitness of GA parameters optimization processing is stable gradually, thePSO fitness changes is insecure dramatically.