基于随机森林的第三次国土调查可行性分析——以水体为例

Feasibility Analysis of Third National Land SurveyBased on Random Forest:Take Water Body as an Example

全国第三次国土调查的一个重要步骤为内业勾绘图斑,需要大量的人力、物力及财力来支撑。利用目视解译的方法进行内业图斑的勾绘,不仅费时费力,而且需要作业人员具有一定的遥感图像解译经验从而满足图斑勾绘的进度及精度。因而,针对全国第三次国土调查,本文基于随机森林的分类方法探讨遥感图像分类的精度是否可以满足全国第三次国土调查图斑勾绘的精度,主要以水体为例。通过利用目视解译的成果来验证随机森林分类的精度,结果表明,随机森林分类的总体精度为91.45%,Kappa系数为0.66,因而,利用随机森林的分类方法可以满足全国第三次国土调查的精度,并且可以精度较高地提取遥感图像中的水体。

One of the important steps of the third national land survey is to draw polygons for the indoor work,which needs a lot of human,material,and financial resources to support.It is time-consuming and laborious to use the method of visual interpretation to draw the polygons and the operators need some experience of remote sensing image interpretation to meet the progress and accuracy of the polygon drawing.Therefore,for the third national land survey,this paper discusses whether the accuracy of remote sensing image classification can meet the accuracy of the third national land survey based on the classification method of random forest,mainly taking water body as an example.By using visual interpretation results to verify the accuracy of the random forest classification,the results show that the overall accuracy of the random forest classification is 91.45%,Kappa coefficient is 0.66,therefore,the use of the random forest classification method can meet the accuracy of the third national land survey,and can extract water from the remote sensing image with high accuracy.