华南地质2025，Vol.41Issue(2)：263-276,14.DOI:10.3969/j.issn.2097-0013.2025.02.002
基于深度学习的湖北秭归县滑坡识别及形变分析
Landslide Identification and Deformation Analysis in Zigui County,Hubei Province Based on Deep Learning
摘要
Abstract
In recent years,landslide disasters have occurred frequently in Zigui County,Hubei Province,posing a significant threat to the production and life of local people.Identifying potential landslides and analyzing their deformation characteristics can help to respond and reduce disaster losses quickly.In this paper,deep learning is carried out using the Landslide4Sens dataset to identify landslides in Zigui County,and time-series InSAR technology is combined to monitor the deformation of slopes.The results show that:(1)The recognition accuracy in the U-Net neural network training results,is 80％,the recall rate reaches 60％,and the F1-value reaches 70％,indicating that the model training effect is good.(2)Based on the construction of the Landslide4Sense dataset and feature sets,landslides can be identified more accurately,providing technical support for large-scale landslide identification.(3)There are relatively more landslides identified in the northern and eastern parts of Zigui County.The landslides are mostly distributed along both sides of the Yangtze River,with the right bank being more than the left bank.There are landslides aggregated around the Three Gorges Dam.When the distance from the dam increases,the number of landslides shows a trend of increasing and then decreasing.The number of landslides is the highest within 20 to 30 kilometers from the dam.(4)In elevation,slope,Normalised Vegetation Index,Normalised Difference Moisture Index,Normalized Brightness Temperature Index,and Bare Soil Index,the number of landslides first increases and then decreases with the index increases.(5)The deformation rate range in Zigui County is from-20 to+20 cm/yr.And the local subsidence areas are mainly distributed in the south of Zigui County,with a subsidence rate of-5 to-15 cm/yr;the subsidence rate in the Yangtze River and tributary areas is-5 to-10 cm/yr,and the deformation gradient is large,with significant subsidence at the confluence.(6)The landslide deformation curve shows a sliding trend,indicating that the landslide identification is reliable.There are stage characteristics of landslide sliding,with March-August being in a sliding trend,and intensifying from April to May,which requires key prevention,and it is relatively stable from September to February.关键词
Landslide4Sense/U-Net神经网络/时序InSAR/滑坡/形变曲线/秭归县Key words
Landslide4Sense/U-Net neural network/time-series InSAR/landslide/deformation curve/Zigui County分类
天文与地球科学引用本文复制引用
殷宗敏,刘学浩,何文熹..基于深度学习的湖北秭归县滑坡识别及形变分析[J].华南地质,2025,41(2):263-276,14.基金项目
国家自然科学基金项目(42107485) （42107485）
中国地质调查局地质调查项目(DD20230104) （DD20230104）
