遥感与地理信息系统学报

抽象的

Cadastral Boundary Extraction and Image Classification Using OBIA and Machine Learning for National Land Records Modernization Programme in India

Thakur V, Doja MN, Ahmad T, Rawat R

The present work is based on the dynamic approach for the extraction of cadastral boundaries and image classification using machine learning algorithms. The efforts are focused on easing the map digitization process in the country. The Large Scale Mean Shift Segmentation algorithm was used for the delineation of cadastral boundaries from two different types of study regions taken up for study, based on their landforms-hills and plains. The quality of segmentation was measured by AssesSeg software. Models using classifiers-Random Forest and Support Vector Machines were trained and their efficiency was tested on multiple images. The behavior of models was observed based on the landforms. The error matrices were generated based on the reference data. We tested these models as demonstrator for updating old maps through image analysis and on the basis of their performance, considered the potential of using them to update land records data in the country. This research shows the possibility of adapting the supervised machine learning methods for the extraction and classification of geographical features using satellite imagery.