[Objective]Quaternary soil is the main source of soil landslide and its distribution and thickness are the important ba-sis for identifying the hidden danger of soil landslide.With the rise of machine learning,the combination of image classification and artificial intelligence algorithm has become the mainstream of remote sensing recognition.[Methods]In this paper,the Zigui syncline Basin of the Three Gorges Reservoir is taken as the research area,the Landsat-8 remote sensing image is used as the basic data source,the existing soil landslide data in the area is used to construct samples,and the machine learning software EnMAP-Box is used to establish the random forest classification model of the quaternary thickness and spatial distribution infor-mation,and screen out the optimal feature subset for identifying the Quaternary soil thickness.The relative thickness spatial distribution of Quaternary system is obtained.[Results]The result show that the spectral characteristics,principal components,vegetation index,humidity,slope,greenness and mean value of Landsat-8 remote sensing images are strongly correlated with the thickness of Quaternary soil,which can be used as an important characteristic factor to identify the thickness of Quaternary soil.The random forest model can effectively identify the soil thickness information of Quaternary system,and the extraction accuracy of rocky area is high.The result of field investigation show that the model has balanced performance and reasonable prediction result,which can be used to identify the Quaternary system in multi-vegetation middle-low mountain environment.[Conclusion]The research result can provide important data support for soil landslide hidden danger identification and risk prevention and control.
关键词
第四系土体/滑坡/相对厚度/机器学习/空间信息提取/三峡库首
Key words
Quaternary soil/landslide/relative thickness/machine learning/spatial information extraction/head area of the Three Gorges Reservoir
维普
万方数据