Revising cadastral data on land boundaries using deep learning in image-based mapping

Bujar Fetai (Author), Dejan Grigillo (Author), Anka Lisec (Author)

Abstract

Geospatial data and information within contemporary land administration systems are fundamental to manage the territory adequately. 3D land administration systems, often addressed as 3D cadastre, promise several benefits, particularly in managing todays complex built environment, but these are currently still non-existent in their full capacity. The development of any complex information and administration system, such as a land administration system, is time-consuming and costly, particularly during the phase of evaluation and testing. In this regard, the process of implementing such systems may benefit from using synthetic data. In this study, the method for simulating the 3D cadastral dataset is presented and discussed. The dataset is generated using a procedural modelling method, referenced to real cadastral data for the Slovenian territory and stored in a spatial database management system (DBMS) that supports storage of 3D spatial data. Spatial queries, related to 3D cadastral data management, are used to evaluate the database performance and storage characteristics, and 3D visualisation options. The results of the study show that the method is feasible for the simulation of large-scale 3D cadastral datasets. Using the developed spatial queries and their performance analysis, we demonstrate the importance of the simulated dataset for developing efficient 3D cadastral data management processes.

Keywords

geodezija;zemljišče;vidna meja;kataster;vzdrževanje;UAV;globoko učenje;geodesy;land;visible boundary;cadastre;maintenance;deep learning;

Data

Language:	English
Year of publishing:	2022
Typology:	1.01 - Original Scientific Article
Organization:	UL FGG - Faculty of Civil and Geodetic Engineering
UDC:	528.4
COBISS:	107071235
ISSN:	2220-9964
Views:	268
Downloads:	86
Average score:	0 (0 votes)
Metadata:

Other data

Secondary language:	Slovenian
Secondary abstract:	Med velike izzive na področju zemljiške administracije v katastrsko razvitih državah štejemo zagotovo vprašanje, kako posodabljate katastrske podatke. Cilj te raziskave je bi razviti pristop za samodejno prepoznavanje vidnih zemljiških mej in preveriti ažurnost katastrskih podatkov o poteku mej z uporabo globokega učenja. Konvolucijsko nevronsko mrežo CNN, ki temelji na nekoliko spremenjeni arhitekturi, smo učili s pomočjo Berkeley segmentacijskega podatkovnega niza 500 (BSDS500), ki je na voljo na spletu. Ta podatkovni niz je namenjen predvsem učenju nevronskih mrež za detekcijo robov in mej. Predlagan model smo testirali na dveh ruralnih območjih v Sloveniji. Na podlagi rezultatov študije smo ugotovili, da so metode za samodejno zaznavanje mej in robov na slikah hitrejše, ko je model enkrat naučen, so pa taki pristopi manj točni v primerjavi z ročnim zajemom robov oziroma mej. Vsekakor so samodejni postopki za zajem podatkov o mejah z emljišč na podlagi slikovnih podob primerni za hiter pregled katastrskih podatkov na geodetskih oziroma katastrskih upravah, to še posebej velja tudi za države z že razvitim katastrom.
Secondary keywords:	geodezija;zemljišče;vidna meja;kataster;vzdrževanje;UAV;globoko učenje;
Type (COBISS):	Article
Pages:	17 str.
Volume:	ǂLetn. ǂ11
Issue:	ǂšt. ǂ5, art. 298
Chronology:	Maj 2022
DOI:	10.3390/ijgi11050298
ID:	15504056