决策树方法在遥感地质填图中的应用

决策树理论在遥感分类中, 分类准确、高效.依据其理论方法, 对青海省民和地区的遥感数据———ETM + (enhanced thematic mapper plus) 进行了分类, 选用的ETM +数据为1999年10月份数据, 数字高程(DEM) 数据来自于1:2 5万民和幅地形图, 数据格式为MapInfo通用格式MIF, 数据进行了坐标转换(地理坐标), 对原始数据进行了处理, 从等高线中提取数字高程.对遥感数据进行地形及光照矫正, 计算植被因子及缨帽变换的3个分量, 同其他5个遥感波段结合形成原始分类图层, 同时确定目标分类结果.原始数据的采样基于目视, 首先采用不同的彩色合成方案突出不同的目标地物, 交互式进行采样, 使用IDL语言编制程序从原始数据中提取地物数字信息, 使用Clementine7.2对数据进行处理, 其中10 %的采样数据验证模型准确率, 其余数据用来推算模型, 对数据进行10次迭代, 同时给予75 %的剪枝, 得到区分不同地物(如红层、黄土等) 的最合适图层(band 1 & band 3)和具体数值, 形成决策树模型, 将决策树模型导入Envi4.0中, 对原始数据(9个图层) 进行计算形成初步分类结果图, 对初步分类结果图进行一定的碎片合并, 最终形成分类结果图.该图同1:2 5万地质图进行对比确认分类的效果, 同传统分类图比较确认决策树分类方法优于传统分类.另外来自于决策树所提取的信息, 有利于地学知识的归纳总结 

Application of Decision Tree Method in Remote Sensing Geological Mapping

The decision tree is an effective and accurate method in remote sensing classification. We use this method to classify the remote sensing data-ETM+ (enhanced thematic mapper plus), which covers most of the area of Minhe County, Qinghai Province. The acquisition date of ETM+ is October 29, 1999. We get digital elevation model (DEM) data from (1:) 250 000 topography map of Minhe area. The format of DEM is MapInfo exchange format *.mif which converted to (geography) coordinate. After primary treatment of the raw data, the DEM data is derived from the contour line. The ETM+ scenes are rectified using the DEM and sun-illumination model. NDVI and other three indexes from Tasseled Cap transform were calculated from RS data. All these indexes are stacked with five RS bands. The target objects are selected. The sampling of the target object is based on visual observation. First, false color composed imagery is essential, and the sampling process is interactive. The digital information of the target object is derived from the program, which compiled by IDL. The decision tree model was calculated by Clementine7.2 software suite. About 10% raw data were used to validate the accuracy of the model. Meantime others were used to build the model. Ten iterative numbers and 75% trim radio are the suit parameters for this model. Then we get the most suitable layer and numerical value for distinguishing different target objects. For instance, distinguishing Tertiary red clastic and loess's best layer is band 1 and band 3. In the next step, we import the model to Envi 4.0 and classify the raw data into different target objects. After some basic treatments, for example clump and (assign) class color, we get the final result map. The map is contrasted with 1:250 000 geology map of Minhe area and achieved the accuracy of classification. The result is that the decision tree method is better than traditional classify methods. Another conclusion is that the rules from decision tree could help geologists to gain a appropriate geological conclusion.