基于机器学习的Landsat数据地层信息提取——以西南天山柯坪地区为例

鲍宽乐; 许文波; 王庆同

引用本文:

【打印本页】【下载PDF全文】【查看/发表评论】【下载PDF阅读器】【关闭】

←前一篇|后一篇→

过刊浏览高级检索

本文已被：浏览 133次下载 459次	码上扫一扫！
分享到：微信更多字体:加大+\|默认\|缩小-
基于机器学习的Landsat数据地层信息提取——以西南天山柯坪地区为例
鲍宽乐¹, 许文波², 王庆同¹
1.中国地质调查局烟台海岸带地质调查中心, 山东烟台 264004;2.电子科技大学资源与环境学院, 四川成都 611731

摘要:

当前机器学习方法不断创新发展,为遥感数据的分析利用提供了很好的平台。结合西南天山柯坪地区沉积岩的典型地质特征,针对1:5万区域地质调查数据,利用Landsat 8数据9个波段的遥感信息进行机器学习方法解译。为增强机器学习过程中参与变量数目,在原始9波段数据的基础上分别采用比值法增强方式、主成分分析法增强方式进行数据叠加。为减弱地质体内部纹理信息,同时不影响地质体之间的边界,笔者采用双边滤波的形式对遥感数据进行进一步处理。选用的极限随机树方法、直方梯度增强随机树、随机森林3种机器学习方法整体识别精度均超过93%,尤其是极限随机树方法达到94.18%。本研究方法可用在其他地质信息解译、地质填图中,值得推广。

关键词: 双边滤波主成分分析波段比随机森林极限随机树直方梯度增强随机树 Landsat 8

DOI：10.12097/j.issn.1671-2552.2023.04.012

分类号:P539.4;P627

基金项目:中国地质调查局项目《胶东北部海岸带与岛礁综合地质调查》(编号:ZD20220604)、《胶东北部海岸带与岛礁综合地质调查》(编号:DD20220990)、《西南天山柯坪地区六幅地质矿产综合调查》(编号:12120115041201)

Stratigraphic information extraction from landsat data based on machine learning—a case study of the Keping area of southwest Tianshan Mountains

BAO Kuanle¹, XU Wenbo², WANG Qingtong¹

1.Yantai Coastal Zone Geological Survey Center,China Geological Survey,Yantai 264004,Shandong,China;2.University of Electronic Science and Technology of China,Chengdu 611731,Sichuan,China

Abstract:

In the current era of artificial intelligence,machine learning methods are constantly innovating and developing.It provides a good platform for effectively analyzing and utilizing of remote sensing data.Combined with the typical geological characteristics of sedimentary rocks in Keping area of southwest Tianshan Mountains,in this paper,remote sensing data from 9 bands of Landsat 8 data were used for machine learning interpretation.In order to enhance the number of variables involved in the process of machine learning,ratio method and principal component analysis method were used to enhance the data superposition based on the original 9-band data.In order to weaken the internal texture information of geological bodies without affecting the boundary between geological bodies,bilateral filtering was used to further process remote sensing data.Three machine learning methods,Extra Trees,Hist Gradient Boosting and Random Forest,were selected.Through experiments,the overall recognition accuracy exceeded 93%,especially 94.18% for the Extra Trees method.It is worth popularizing the research method of this paper in other geological information interpretations and geological mapping.

Key words: bilateral filter principal component analysis band ratio Extra Trees Random Forest Hist Gradient Boosting Landsat 8