摘要
传统的土地调查摸排类工作通常是基层调查人员采用区域巡查或基于卫星遥感影像人工提取变化图斑下发的方式开展,周期长且成本高。为缩短工期、降低成本,采用深度学习的方法训练房屋建筑样本,自动提取耕地范围内房屋建筑,并以人工检核的方式进行补充,建立以月度为周期的农村乱占耕地建房调查体系。试验结果表明,深度提取耕地范围内房屋建筑物的正确率为76.2%,召回率为82.4%,可极大地提高工作效率,为后续土地摸排类工作提供借鉴。
The traditional land survey and inventory work is usually carried out by the grassroots investigators in the form of regional patrolling or manual extraction of change patches based on satellite remote sensing images,which has a long cycle and high labor cost.The method of deep learning is used to train the sample of housing construction,automatically extract the housing construction within the scope of cultivated land,and supplemented by manual inspection,so as to establish a monthly survey system of housing construction in rural illegally occupied cultivated land.The experimental results show that the accuracy rate of deep learning extraction of houses and buildings within the scope of cultivated land is 76.2%,and the recall rate is 82.4%,which can greatly improve the work efficiency and provide a reference for the follow-up land inventory work.
作者
马庆伟
姚超峰
郭长恩
MA Qingwei;YAO Chaofeng;GUO Chang′en(No.801 Brigade of Hydrogeology and Engineering Geology of Shandong Provincial Bureau of Geology&Mineral Resources(Shandong Provincial Geo-mineral Engineering Exploration Institute),Jinan 250000,China;Shandong Provincial Engineering Technology Research Center for Groundwater Environmental Protection and Remediation,Jinan 250000,China)
出处
《测绘与空间地理信息》
2024年第8期155-158,共4页
Geomatics & Spatial Information Technology
关键词
农村乱占耕地建房
深度学习
卫星遥感影像
常态化调查
building houses in rural illegally occupied cultivated land
deep learning
satellite remote sensing images
normalization survey
