基于决策树的Landsat多光谱影像分类方法  

Classification Methods of Landsat Multi-spectral Images Based on Decision Tree

作　　者：黄立贤[1] 沈志学[1] 

机构地区：[1]中国工程物理研究院流体物理研究所,四川绵阳621900

出　　处：《光电技术应用》2011年第3期49-52,共4页Electro-Optic Technology Application

摘　　要：构建了一种结合光谱特征和纹理特征的多光谱影像决策树分类方法。以Landsat-7影像作为试验数据,通过分析Landsat-7影像的光谱特征值及NDVI、NDWI和NDBI特征值,确定各类地物的综合阈值,同时运用灰度共生矩阵对影像进行纹理信息提取,得到对比度、熵、逆差矩和相关性等纹理特征图像。在此基础上,运用决策树分类法对Landsat-7影像进行分类。结果表明,结合光谱特征和纹理特征的决策树分类方法,相比传统的最大似然法和决策树法,具有更高的分类精度。A classification method of decision tree for multi-spectral images, which combines the spectral features with the texture features is presented. The Landsat-7 images are taken as the experiment data, the integrated thresholds of the various ground objects are determined by analyzing characteristic values of the Landsat-7 images spectrum, NDVI, NDWI and NDBI. Meanwhile, the texture information of images including the contrast, entropy, homogeneity and correlation are acquired by means of using Gray Level Co-occurrence Matrices. On this basis the Landsat-7 images are classified by the decision tree classification method. The results indicate that the accuracy of decision tree classification based on spectral and texture features is higher than that of the maximum likelihood and traditional decision tree method.

关 键 词：多光谱影像 决策树分类 纹理特征 光谱特征 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置] P237[自动化与计算机技术—控制科学与工程]