摘要
选择昆明市作为研究区,以2011年Landsat TM影像为基础数据,通过分析研究区地形特征,提出把研究区进行分区并分别确定高程、坡度决策规则的改进型决策树分类方法,并结合分析的光谱特征规律,在决策分类中引进了比值型指数、NDVI值,构建基于光谱特征和地学辅助知识的决策树信息提取模型,最后对传统计算机自动监督分类方法与决策树信息提取模型方法解译的昆明市土地利用数据的精度进行评价。研究结果表明:基于改进的决策树分类方法进行遥感信息提取的昆明市土地利用数据的Kappa指数比传统监督分类方法提高了0.234,分类精度提高了17.03%;从各种地类类型的测试样本点平均正确率来看,改进的决策树分类方法比传统监督分类方法提高了21%,大大提高了Landsat TM遥感数据分类的精确度和可靠性。
Kunming was chosen as the study area and terrain features of study area were analyzed based on the Landsat TM image data in 2011. The improved decision tree classification method was put forward, which deter- mined the decision rules of elevation and slope respectively through study area partition. In addition, ratio index and NDVI value participation decision and classification, and structures decision tree information extraction mod- el was introduced based on the spectral characteristics and learning aids knowledge. Finally, accuracy of Kun- ming city land use data was interpreted by traditional computer automatic classification method and decision tree information extraction model was valued. The results show that Kappa index and classification precision, ob- tained from remote sensing information extraction using decision tree classification method, increased by 0. 2343 and 17.03% compared with traditional supervised classification method. From various types of land, tests sam- ple average correct rate improved by 21% obtained from decision tree classification method relative to supervised classification method. Obviously, the improved decision tree classification method greatly improves accuracy and reliability of Landsat TM remote sensing data classification.
出处
《贵州大学学报（自然科学版）》
2013年第6期39-46,52,共9页
Journal of Guizhou University:Natural Sciences
基金
国家自然科学基金(31060114)
关键词
土地利用
地学信息
光谱特征
决策树
监督分类
评价
landuse
geological information
spectral characteristics
decision tree
supervised classification
evaluation
