基于决策树方法的云南省森林分类研究

森林分类对于理解森林生态系统结构和功能具有重要意义。由于云南省地形和森林类型复杂,首先按云南省的16个行政区划将全省Landsat TM影像分为对应的16个区域。以TM波段1~5和7,以及由植被指数、缨帽变换、主成分变换、DEM组成的18个变量组,统计训练样本光谱值均值变化和光谱值与频率间的关系。利用交点计算公式计算类间最佳分类界点进而建立决策树,逐一分离各区的所有森林类型,将分类结果合并得到云南省阔叶林、针叶林和针阔混交林类型分布图。最后将分类结果与监督分类中的最大似然比法的分类结果进行对比。结果表明:监督分类的总体分类精度为74.39%,Kappa系数为0.63,决策树方法的总体分类精度为86.61%,Kappa系数为0.80,说明决策树方法可以提取高精度的云南省森林类型,进而为该区域森林叶面积指数和生物量反演等研究提供基础数据支持。

Study on Forest Classification in Yunnanbased on Decision Tree Algorithm

Forest classification plays an important role in understanding the structure and function of forest ecosystem.Because the type of topography and forest richness were complex,the Landsat TM images was divided into 16 corresponding pieces according to administrative regionalization in Yunnan province.Firstly,based on 18 variations,such as TM 1~5 and 7,and vegetation index,KT transform results,Principal Component(PC) transform results,DEM data and so on,this study computed the mean variation of spectral values of training sets and the relationship between the spectral values and its frequency in the paper.Secondly,using intersection point algorithm,the thresholds of each classification were obtained and decision tree was established to separate each class step by step.Thirdly,by merging the classification results of 16 pieces into one large area,the total classification result of Yunnan forest was achieved.Finally,we selected Shangri|la County as a validation case,and the results derived by decision tree method,which were compared with those of maximum likelihood approach.It shows that the total accuracy of the former is 86.61% (Kappa=0.80),while the latter is 74.39% (Kappa=0.63).The result indicates that the decision tree method is suitable and effective for forest classification and can provide basic data support for the study of forest LAI and biomass estimation in Yunnan province.