基于改进决策树分类算法的遥感影像分类研究

针对现有分类器对遥感影像分类结果存不准确的问题,本文提出了一种基于决策树分类器的遥感影像分类方法,该方法以复合决策树Boost Tree思想为基础,首先利用分形理论中的毯模型提取遥感影像的纹理特征,根据遥感影像分类的特点,构造新的单棵决策树生成算法对遥感影像进行分类。以北京市五环内区域为研究区,使用landsat7 ETM数据源,实现了基于分形纹理特征、光谱特征的改进决策树分类。实验结果表明：通过毯模型提取的纹理特征可以很好地表达表面特征,辅以该纹理信息的改进决策树分类精度相比于只用光谱信息进行分类的精度有一定的提高,改善了分类效果。

Research on Remote Sensing image Classification based on Improved Decision Tree Classification Algorithm

According to the classification of remote sensing images result in inaccurate classifier existing problems, this paper proposes a method for classification of remote sensing image vegetation based on decision tree classifier, this method is based on the composite decision tree Boost Tree, texture feature firstly extracted using remote sensing images blanket model in fractal theory, according to the characteristics of remote sensing image the classification, the construction of a new single decision tree algorithm to classify the remote sensing image. The improved decision tree classification based on fractal texture features and spectral features is realized by using the Landsat7 ETM data source as the research area in the five ring area of Beijing. The experimental results show that the texture features extracted from carpet models can express the surface features very well, and the classification accuracy of the improved decision tree supplemented with the texture information is improved compared with the spectral information only, and the classification effect is improved.