摘要
提出在多维特征空间中以互信息为评价指标进行特征选择,在特征子集中应用支持向量机(SVM)分类器实现图像监督分类的方法。首先提取图像的光谱、纹理和颜色特征,得到多特征的高维特征空间,然后用最大相关和最小冗余的互信息作为评价标准,用10-fold交叉验证误差率选择特征子集,最后用基于径向基函数的SVM实现图像的分类。实验表明,该方法能明显提高图像分类的精度。
In this paper,a novel remote sensing image classification method was proposed,which was based on feature selection and Support Vector Machine （SVM） classifier. A minimal redundancy and maximal relevance criterion based on mutual information was applied to selection a set of informative and non-redundant Gabor texture feature, HSV color feature and spectral feature from high spectral remote sensing image, which are then further enhanced by SVM based on radial basis function for supervise classification. Experimental results show this method leads to promising improvement on classification accuracy compare with other traditional ones.
出处
《地理与地理信息科学》
CSCD
北大核心
2009年第3期19-22,41,共5页
Geography and Geo-Information Science
基金
国家重点基础研究发展计划项目(2006CB701300)
中南林业科技大学青年科学基金项目(07042B)
关键词
图像分类
互信息
特征选择
SVM
image classification
mutual information
feature selection
SVM
