摘要:
目前的遥感影像分类研究中,决策树的生成完全依赖于现有的数据挖掘软件,缺少对决策树算法的深入研究和改进。本文以遥感影像分类为背景,采用BoostTree算法作为模型,通过算法改进构建了一种新的复合决策树算法———AdaTree,并以该算法为基础,设计实现了决策树遥感影像分类系统。以AdaTree算法作为分类器,分别对Landsat7ETM+影像和WordView2影像进行了基于像元和面向对象的分类实验,并与BoostTree和SVM算法进行了比较。实验结果表明,AdaTree算法在分类精度上要优于BoostTree和SVM算法,平均Kappa系数分别达到0.905 2和0.939 8。
Abstract:
As one of main classification methods used in data mining,the decision tree algo-rithm is widely used in remote sensing image classification.However,in current studies ofremote sensing image classification,the building of decision trees was found to be dependenton existing data mining software,with little research work focused on decision tree algo-rithms.Based on the BoostTree algorithm,we propose a new algorithm of decision tree en-sembles for remote sensing image classification-AdaTree which is a combination of C4.5andAdaBoost.M1algorithms.In AdaTree,the structure of C4.5and the final hypothesis of Ad-aBoost.M1were modified.With the AdaTree classifier algorithm,apiece of software wasdeveloped for cell-based and object-oriented remote sensing image classification.An experi-ment with Landsat7ETM+ and Wordview2images showed accuracy and efficient improve-ments of the AdaTree classifier when compared with BoostTree and SVM,either in cells-based or object-oriented classification.Its average Kappa coefficients reached 0.905 2and0.939 8.
下载:
