摘要
地裂缝需要被持续监测,但是地裂缝探测仍需人工实地勘探,因此自动地裂缝提取具有重要意义。为此,该文提出一种深度学习模型,利用无人机影像自动提取地裂缝,该模型针对地裂缝相对其他地物具有细长结构的特征,设计了地裂缝提取网络;针对人工准确标注地裂缝蜿蜒曲折的形态费时费力等特点,设计了一种弱监督的方法对人工标签进行优化,改善人工标签不准确的情况。利用朔州市平鲁区无人机影像验证方法有效性,实验结果表明:地裂缝能被有效提取,召回率达91.4%,并利用提取的地裂缝生成了地图产品,可用于辅助区域内地裂缝风险警示和成因分析。
Ground fissures need to be continuously monitored,but ground fissure detection still needs manual field exploration,so automatic extraction of ground fissures is of great significance.A deep learning model is proposed to automatically extract ground fissures from UAV images.In this model,a specific network is designed for the elongated ground fissures.Because manually labeling ground fissures is slow and laborious,a weakly supervised learning method is proposed to optimize the artificial label and improve the inaccuracy of the artificial label.The effectiveness of the model is verified by using the UAV images in Shuozhou Pinglu District.The experimental results show that the ground fissures can be extracted effectively,and the recall of them reaches 96.8%.A map product is generated by the extracted ground fissures,which can be used to assist the risk warning and genetic analysis of ground fractures in the region.
作者
王臻
王辉
李国锋
WANG Zhen;WANG Hui;LI Guofeng(College of Land Science and Technology,China University of Geosciences(Beijing),Beijing 100083,China;Research Center of Big Data Technology,Nanhu Laboratory,Jiaxing 314000,China;Coal Geological Geophysical Exploration Surveying&Mapping Institude of Shanxi Province,Jinzhong 030600,China)
出处
《实验技术与管理》
CAS
北大核心
2022年第3期51-56,共6页
Experimental Technology and Management
基金
国家自然科学基金青年科学基金项目(41901414)。
关键词
地裂缝
无人机影像
深度学习
弱监督学习
ground fissures
UAV images
deep learning
weakly supervised learning
