支持向量机和BP网络改进模型的性能对比研究

Comparative Study on Support Vector Machine and Improved BP Network Models

通过引入支持向量机(SVM)方法,提出了基于SVM的遥感图像多类分类模型,分析了SVM多类分类器的构造及其参数选取问题,并结合实例,讨论了SVM分类器性能随其本身参数变化情况,最后与几种代表性的BP网络改进模型进行了系统的对比分析。实验表明,SVM方法的分类时间要远大于改进的BP模型,而分类精度优于BP网络改进模型中效果最好的几种优化算法3个百分点左右,是一种有效的图像分类方法。

The paper imports Support Vector Machine （SVM）, and proposes SVM-based multi-class classified model for remote sensing image, and analyzes construction and parameter choice problem of SVM classifier. At last taking remote sensing image classification as samples, change situation of SVM along with itself parameter is discussed, and comparative study on convergent velocity and classified effect is done among SVM and several improved BP network models. Experimental result suggests that classified consumed time of SVM is far longer than improved BP model and classified precision preponderate over the best improved BP models about 3 percent. It is a kind of efficient image classification method.