一种面向对象的CART决策树火烧迹地提取方法

An object-oriented method for extracting burned area using CART decision tree

现有的火烧迹地遥感提取主要侧重于对光谱信息的判识,对遥感影像的形状、纹理、空间上下文等特征的挖掘尚不充分。为此,本文提出了一种面向对象的分类回归树算法(CART)决策树火烧迹地提取方法,旨在提升火烧迹地遥感信息提取的精度和可靠性。为验证方法的可行性,本文选取四川省冕宁县“4·20”森林火灾为研究区,以国产高分一号B星(GF-1B)卫星数据为数据源,对研究区影像进行面向对象的最优尺度分割,并采用CART决策树算法,根据不同地物的光谱、形状和纹理特征从中自动获取最优特征及其阈值,构建决策树实现火烧迹地提取。结果表明:该方法在火烧迹地上的提取精度(总体精度92.00%)和可靠性(Kappa系数85.56%)均优于既有的监督分类技术方法。相关研究方法和实验结果可为火烧迹地精准提取与灾后评估等研究提供参考。

The existing remote sensing extraction of burned area mainly focuses on the discernment of spectral information,and the mining of features such as shape,texture and spatial context of remote sensing images is not sufficient.To this end,this paper proposed an object-oriented classification and regression tree(CART)decision tree burned area extraction method,which aimed to improve the accuracy and reliability of remote sensing information extraction of burned area.To verify the feasibility of the method,this paper selected the 4.20 forest fire in Mianning County,Sichuan province as the study area,took the domestic GF-1B satellite data as the data source,carried out the object-oriented optimal scale segmentation of the study area image,and adopted the CART decision tree algorithm to automatically obtain the optimal features and their thresholds from them according to the spectral,shape and texture features of different features,and constructed a decision tree to realize the extraction of burned area.The result showed that the extraction accuracy(overall accuracy 92.00%)and reliability(Kappa coefficient 85.56%)of this method on burned area were better than the established supervised classification technique methods.The research method and experimental result could be used as a reference for accurate extraction and post-disaster assessment of burned area.