基于支持向量机的几种核函数遥感图像分类比较

Comparison of Remote Sensing Image Classification by Several Kinds of Support Vector Machine Kernel Function

图像分类是人们获取信息的一种重要的手段,传统的分类方法是以经验风险最小化为归纳原则,只有当训练样本数趋于无穷时,其性能才能达到理论上的最优。当样本不足时,传统的分类方法往往不能达到理想的分类精度。与传统的人工神经网络相比,支持向量机理论体现了结构风险最小化原则,它不仅结构简单,泛化能力强,而且能较好的解决小样本、高维数据和局部极小等实际问题。本文以试验区的地物分类为研究背景,建立了支持向量机的算法框架,并分别使用多项式核函数,径向基核函数、Sigmoid核函数以及线性核函数四种核函数对图像进行了多类别分类实验。

Image classification is an important means for people to obtain information, the traditional classification methods are based on empirical risk minimization principle, only when the number of training samples is infinite, its performance can reach the optimal theory. When the sample is insufficient, the traditional classification methods often fail to achieve the desired accuracy of classification. Compared with the traditional neural network, support vector machine theory embodies the structural risk minimization principle, it not only has the advantages of simple structure, strong generalization ability, and can solve the small sample, high dimension data and local minimum problems. Based on the classification test area as the research background, the algorithm of support vector machine is built, and the use of polynomial kernel function , radial basis function, Sigmoid kernel and the linear kernel function four kinds of kernel function to do image classification experiment.