Cadastral boundary delineation using deep learning and remote sensing imagery: state of the art and future developments

Global estimates indicate that 70-75% of the people worldwide do not have access to a legal land administration system. The absence of such a system has a negative impact on essential developments such as tenure security, agricultural productivity, and sustainable development. Therefore, the UN included the formal registration of property rights, ownership, and value in several targets of the sustainable development goals. A concept that tries to speed up the registration of legal land rights is fit-for-purpose land administration. Leading principles in this approach are the use of visible cadastral boundaries and the delineation of these boundaries by using remote sensing imagery. In recent years, these principles have resulted in a growing demand for automated cadastral boundary extraction methods. Recent studies on cadastral boundary extraction using deep learning and remote sensing imagery look promising. However, the number of studies is limited. Recent studies are based on small study areas and small data sets, did not implement state-of-the-art deep learning models, and did not investigate the transferability and generalization ability of the models to other geographical locations. This paper reviews the applications of deep learning and remote sensing imagery for cadastral mapping and describes several research gaps. Possible improvements include building benchmark datasets, evaluating the proportion of visible/invisible boundaries in these benchmark datasets by analyzing the overlap between cadastral and topographical boundaries, applying deep learning applications from other remote sensing fields for cadastral mapping, and creating human-in-the-loop deep learning pipelines.