Deep Fully Convolutional Networks for Cadastral Boundary Detection from UAV Images

Abstract

There is a growing demand for cheap and fast cadastral mapping methods to face the challenge of 70% global unregistered land rights. As traditional on-site field surveying is time-consuming and labor intensive, imagery-based cadastral mapping has in recent years been advocated by fit-for-purpose (FFP) land administration. However, owing to the semantic gap between the high-level cadastral boundary concept and low-level visual cues in the imagery, improving the accuracy of automatic boundary delineation remains a major challenge. In this research, we use imageries acquired by Unmanned Aerial Vehicles (UAV) to explore the potential of deep Fully Convolutional Networks (FCNs) for cadastral boundary detection in urban and semi-urban areas. We test the performance of FCNs against other state-of-the-art techniques, including Multi-Resolution Segmentation (MRS) and Globalized Probability of Boundary (gPb) in two case study sites in Rwanda. Experimental results show that FCNs outperformed MRS and gPb in both study areas and achieved an average accuracy of 0.79 in precision, 0.37 in recall and 0.50 in F-score. In conclusion, FCNs are able to effectively extract cadastral boundaries, especially when a large proportion of cadastral boundaries are visible. This automated method could minimize manual digitization and reduce field work, thus facilitating the current cadastral mapping and updating practices.

Bridging the Semantic Gap between Land Tenure and EO Data: Conceptual and Methodological Underpinnings for a Geospatially Informed Analysis

“…In light of state-of-the-art methods in land administration, a deep-learning is becoming highly prominent for the detection of cadastral boundaries [12,19]. Recent evidence suggests that deep fully convolutional networks (FCNs) ensures the high accuracy rather than gPb and MRS, with results of 0.79 in precision, 0.37 in recall and 0.50 in F-score [19].…”

“…In light of state-of-the-art methods in land administration, a deep-learning is becoming highly prominent for the detection of cadastral boundaries [12,19]. Recent evidence suggests that deep fully convolutional networks (FCNs) ensures the high accuracy rather than gPb and MRS, with results of 0.79 in precision, 0.37 in recall and 0.50 in F-score [19]. For optimizing image segmentation, one study by [12] not only introduced the interactive boundary delineation workflow, but also examined the better suitability of the deep learning in cadastral mapping with convolutional neural networks (CNNs) by comparing random forest (RF) in machine learning: RF-derived boundary likelihoods (accuracy: 41%, precision: 49%), CNN-derived boundary likelihoods (accuracy: 52%, precision: 76%).…”

“…Convolutional Networks (FCNs) [19] Aerial imagery and UAVs √ Automatic boundary classification: Random Forest (RF), Convolutional Neural Networks (CNN) [12] Remote Sens. 2020, 12, 255 6 of 26…”

“…Literature shows applications ranging from environmental and regional studies to economics, and peace and conflict research, for example [6][7][8][9]. More specific to the interest of this paper, there is growing body of literature on methods to extract and map cadastral boundaries using EO data [10][11][12][13][14][15][16][17][18][19]. However, this literature rarely effectively bridges the knowledge gap between social land tenure and spatial descriptions of boundaries.…”

Delineation of Agricultural Field Boundaries from Sentinel-2 Images Using a Novel Super-Resolution Contour Detector Based on Fully Convolutional Networks

“…For instance, in [14], CNNs are applied to identify slums' degree of deprivation, considering different levels of deprivation including socio-economical aspects. FCN applications include detection of cadastral boundaries [15] and delineation of agricultural fields in smallholder farms [4]. Despite the wide applications of these methods in LCLU classification, their capabilities for boundary delineation have not been explored for medium resolution data such as Sentinel-2; they are limited to VHR images [4].This study introduces a multiple dilation FCN (MD-FCN) for AFB delineation using free (medium resolution) Sentinel-2 images.…”