基于GEE和机器学习的不透水面提取研究——以成渝地区为例  

Research on the impervious surface extraction based on GEE and machine learning:take Chengdu-Chongqing Region as an example

作　　者：夏军 刘洪江 朱林富 段捷 XIA Jun;LIU Hongjiang;ZHU Linfu;DUAN Jie(School of Tourism,Leshan University,Leshan 614000,China)

机构地区：[1]乐山师范学院旅游学院,四川乐山614000

出　　处：《测绘工程》2023年第2期21-29,共9页Engineering of Surveying and Mapping

基　　金：四川省科技计划资助项目(2020YFS0354);四川旅游发展研究中心课题(LY21-19,LY21-22);乐山师范学院科研启动项目(RC210121)。

摘　　要：随着城镇建成区范围不断扩大,不透水面面积急剧扩张,对区域生态和经济产生严重影响,对不透水面的快速、准确识别显得尤为重要。文中以成渝地区双城经济圈的两大核心城市为研究对象,基于GEE云平台和Landsat 8影像,构建光谱波段、光谱指数和纹理指数的分类特征,利用最小距离(MD)、分类回归树(CART)、支持向量机(SVM)、随机森林(RF)和朴素贝叶斯(NB)5种机器学习算法,提取不透水面信息。结果表明,RF算法效果最好,提取结果与实际最相符,CART和SVM算法精度差异不大。本研究可为不透水面提取提供方法和技术参考。With the continuous expansion of the built-up area of cities and towns,the area of impervious surface has expanded rapidly,which has a serious impact on regional ecology and economy.The rapid and accurate identification of impervious surface is particularly important.This paper takes the two core cities in the Chengdu-Chongqing Double City Economic Circle as the study area.Based on the GEE cloud platform and Landsat 8 remote sensing images,the spectral bands,spectral indexs and image textures are constructed and be used as classification features of impervious surfaces,using five machine learning algorithms:Minimum Distance(MD),Classification and Regression Tree(CART),Support Vector Machine(SVM),Random Forest(RF)and Naive Bayes(NB)to extract impervious surface information,the results show that the RF algorithm has the best effect,and the extraction result is the most consistent with the reality.The accuracy of the CART and SVM algorithms is not much different.This research can provide method and technical reference for impervious surface extraction.

关 键 词：双城经济圈 不透水面 机器学习 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]