The Study Of Agricultural Data Classification Based On Support Vector Machine

With the fast development of agricultural information technology,agricultural data accumulated and increased rapidly,more and more demand for agricultural data classification.How to transfer these agricultural data into valuable information quickly and efficiently,is an important topic of the current information and computer technology research.Because of these agricultural data with the characteristics of dynamic,regional,heterogeneity and timeliness,lead to the difficulty of the problems of agricultural classification on the rise.The traditional classification algorithm need to the assumptions of sample size tends to infinity,and many algorithms need to establish accurate mathematical model in dealing with a problem,in reality,these algorithms are often poor performance.Compared with the previous machine learning methods,support vector machine is a universal method of machine learning,it has many advantages in theory and practice,and better solved the problem of nonlinear,dimension disaster,difficult modeling and so on.On the basis of the characteristics of agricultural data,the classification method of support vector machine is proposed with based on analyzing the disadvantage of traditional method considering.It's important to solve the two key problems of agricultural data classification,the not high enough classification accuracy and slow training speed.Puts forward two methods of kernel function improvement,the experiments prove that the two methods are feasible and superior,better improve the learning ability of kernel function classification.The research work and achievements include the following several aspects:1.After descriptions in detail of the basic theories of classification algorithm,research on and analysis of the basic conception,derivation process,advantages and disadvantages of SVM classification algorithm and SMO algorithm,this dissertation takes several comparative experiments of several typical classification algorithms,and does some comparisons and analyses according to all indicators of classifier for evaluating.2.This dissertation takes deep researches in the kernel function of SMO algorithm,an improved method was puts forward,based on the analysis of expected upper bound of classification error rate of the test sample,using the method of reducing the number of the support vector to decrease the rate of classification error rate,and increasing the absolute value of the coefficient of quadratic term to improve the precision of classification.combined with the grid search method to optimize the related parameters of SMO algorithm based on the improved kernel function.The results of experiments show that this improved method can be better overcome the defect of the SMO algorithm for large learning task,the classification accuracy was improved,and the modeling time was greatly reduces.3.After research on the two important types of kernel function,one is the local kernel function,and the other is the global kernel function,this dissertation takes comparative experiments of classification ability of the two types of kernel function.Combined with the characteristics of the two types of kernel function,a hybrid kernel function was constructed which is linearly combined by Polynomial kernel function and RBF kernel function,combined with the grid search method to optimize the related parameters of SMO algorithm based on the hybrid kernel function.The experimental results prove that the mixed kernel function is better than the Polynomial kernel function and the RBF kernel function,not only the classification accuracy is higher,but also compared with the Gaussian kernel function,reduces the complexity of time,save a lot of modeling time.This dissertation has important value and significance for deep researches in the classification methods and theories of agricultural data,The more quickly and accurately methods to deal with the collected agricultural data further improve the research on our country's agriculture and the development of science and technology.