为了应对全球70%的未注册土地权的挑战,对地籍测绘方法的需求不断增长.由于传统的现场实地测量既耗时又耗费人力,因此土地管理部门一直提倡基于遥感的地籍测绘,但基于遥感影像的自动划界的准确性仍然是一项重大挑战.在这项研究中,使用无人机获得的图像来探索深度全卷积网络(Fully Convolu-tional Networks,FCN)在城市和城郊地区进行地籍边界提取的能力.在甘肃天水的两个地点使用其他最先进的技术来测试FCN、多分辨率分割(Multi-Resolution Segmentation,MRS)和全局化边界概率(Globalized Probability of Boundary,gPb)算法的性能.实验结果表明:FCN在两个研究领域的表现均优于MRS和gPb,精度平均为0.79,召回率为0.37,F评分为0.50.总之,FCN能够有效地提取地籍边界,尤其是在大量地籍边界可见的情况下.这种自动化方法可以最大限度地减少手动数字化并减少实地工作,从而促进当前的地籍测绘和更新做法.
Research on Automatic Extraction of Cadastral Boundaries Based on Deep Learning
In order to address the challenge of 70% of unregistered land rights worldwide,there is a growing de-mand for inexpensive and fast cadastral surveying and mapping methods.Due to the time-consuming and labor-intensive nature of traditional on-site measurements,land management departments have always advocated for ca-dastral surveying and mapping based on remote sensing,but the accuracy of automatic demarcation based on re-mote sensing images remains a major challenge.In this study,images obtained by UAVs are used to explore the ability of the Fully Convolutional Network(FCN)extracting cadastral boundaries in urban and suburban areas,and other state-of-the-art technologies are used to test the performance of the FCN,Multi-Resolution Segmentation(MRS),and the Globalized Probability of Boundary(gPb)algorithm in Tianshui and Gansu.Experimental results show that the FCN performs better than MRS and the gPb in the two research fields,with an average accuracy of 0.79,a recall rate of 0.37 and an F-score of 0.50.In summary,the FCN can effectively extract cadastral boundaries,especially when a large number of cadastral boundaries are visible,and this automation method can minimize manual digitization and reduce fieldwork,so as to promote current cadastral surveying and mapping and update methods.
NETL
NSTL
万方数据
