基于决策树的土地利用分类方法研究

Study of Land Use Classification Based on Decision Tree Method

以新疆乌鲁木齐市部分区域为研究区,利用主成分分析法对Spot-5影像进行数据压缩,运用灰度共生矩阵对第一主成份进行纹理信息提取,分析Landsat-7影像的光谱特征值及NDVI和NDBI特征值,确定各类地物的综合阈值,最后运用决策树分类法对Landsat-7影像进行分类。将分类结果与最大然法分类结果相比较,结果表明,决策数分类较最大似然法分类的精度提高了5.66%,Kappa系数提高了7.89%。说明决策树分类能够灵活、有效运用纹理等辅助信息,更好地区分光谱特征相似的目标地物,具有更高的准确性。

In this paper, the partial regions of Urumqi in Xinjiang were taken as research area, principal components were extracted from Spot- 5 image, texture information was acquired by means of using Gray Level Co - occurrence Matrices from the first principal component, then threshold was selected from the characteristic values of the Landsat - 7 image Spectrum, NDVI and NDBI. At last, the decision tree classification was used to classify the landsat- 7 image. The Landsat- 7 image was classified with the decision tree classification method. Then the classification msuhs and maximum likelihood classification were compared with each other. The results indicated that the accuracy of decision tree classification was 5.66% higher than that of the maximum liklihood classification. Kappa coefficient was increased by 7.89%.