摘要
四川地形复杂,山区纵横交错处滑坡具有频发、突发、易发的特点,对人民财产和环境资源造成极大的危害,因此开展滑坡的识别检测,提取相关信息,对滑坡灾害预防监测及灾后预备有着重要的意义。针对传统目视解译方法经济成本高、耗时耗力、历史样本收集困难的问题,引入了高程、坡度、坡向、岩性、地表起伏程度、距断层距离、距水系距离、距道路距离、归一化植被指数9个滑坡影响因子,对历史滑坡的判识中引入影响因子的信息量值进行定量分析,增强了历史滑坡样本数据准确性;其次针对滑坡自动识别结果可能存在的定位不准确、分割边界模糊等问题,采用递归特征金字塔网络和DIoU损失对Mask R-CNN模型进行改进,提出滑坡智能识别改进算法。评价结果表明:改进算法相比原始模型,精确率提高了3.6%,召回率提高5.2%,对四川省青川县历史滑坡进行准确识别与边界分割,识别准确率达74.4%。随着卫星遥感手段与深度学习技术的发展,该改进算法对滑坡智能识别、构建地质灾害风险评价体系提供信息基础与理论参考具有重要意义。
[Objective]The complex topography of Sichuan Province,characterized by intersecting mountainous terrain,leads to frequent,sudden,and highly susceptible landslides.These events pose significant threats to both people's property and environmental resources.Therefore,conducting landslide identification and charaterization are crucial for effective hazard prevention,monitoring,and post-disaster preparedness.[Methods]To overcome the limitations of conventional visual interpretation methods-including high economic costs,time-intensive procedures,labor demands,and challenges in acquiring historical samples,this study incorporates multiple landslide-influencing factors such as elevation,slope gradient,and aspect into the analysis framework.A quantitative information value analysis was conducted to evaluate the predictive capacity of these influencing factors for historical landslide identification,thereby improving the reliability of historical landslide inventories.To solve issues such as inaccurate localization and ambiguous segmentation boundaries in automatic landslide identification results,this paper improves the Mask R-CNN model using a recursive pyramid network and DIoU loss,proposing an improved algorithm for intelligent landslide identification.[Results]Evaluation results demonstrate that the enhanced algorithm significant improvements over the baseline Mask R-CNN,with 3.6%increase in precision and 5.2%increase in recall.The model attains 74.4%identification accuracy in Qingchuan County,Sichuan,showing particular effectiveness in delineating historical landslide boundaries with clear geomorphological fidelity.[Conclusion]Combining satellite remote sensing with deep learning advancements,this improved algorithm enables intelligent landslide identification and supports data-driven risk assessment,offering critical insights for geohazard mitigation.
作者
饶炜博
陈刚
邹崇尧
范小洁
常富强
何建权
林晓静
李显巨
唐骞
RAO Weibo;CHEN Gang;ZOU Chongyao;FAN Xiaojie;CHANG Fuqiang;HE Jianquan;LIN Xiaojing;LI Xianju;TANG Qian(College of Marine Science and Technology,China University of Geosciences(Wuhan),Wuhan 430074,China;Badong National Observation and Research Station of Geohazards,China University of Geosciences(Wuhan),Wuhan 430074,China;School of Computer Science,China University of Geosciences(Wuhan),Wuhan 430074,China;Key Laboratory of Geological Survey and Evolution of Ministry of Education,China University of Geosciences(Wuhan),Wuhan 430074,China;Hubei Institute of Surveying and Mapping,Wuhan 430074,China)
出处
《地质科技通报》
北大核心
2025年第4期48-61,共14页
Bulletin of Geological Science and Technology
基金
湖北省自然资源厅科学研究项目(ZRZY2024KJ03)
湖北巴东地质灾害国家野外科学观测研究站开放基金项目(BNORSG-202415)。
关键词
滑坡识别
数据增强
深度学习
信息量值
滑坡影响因子
改进算法
landslide identification
data enhancement
deep learning
information quantity value
landslideinfluencing factor
improved algorithm
