Scienceline Publication Repository

Scienceline Publication Repository

Scienceline Publication

Mapping Buildings from Semi-Informal Settlements Using Non-Parametric Classifiers: A Case of Old Naledi

(2024) Mapping Buildings from Semi-Informal Settlements Using Non-Parametric Classifiers: A Case of Old Naledi. Journal of Civil Engineering and Urbanism. 149–157. ISSN 2252-0430

[img] Text
JCEU14(3s)149-157,2024.pdf - Published Version

Download (1MB)

Official URL: http://dx.doi.org/10.54203/jceu.2024.14

Abstract

Building footprints are essential for planning and designing new infrastructure like water reticulation, electricity transmission, sewer, and road networks. They are also necessary for delivery, census, and disaster management. It is therefore important to have up-to-date maps and GIS databases for service provision. However, mapping building of footprints in semi-informal settlements is problematic because of the spatial heterogeneity of settlements. This study evaluates three non-parametric machine learning algorithms for extracting building footprints from WorldView-2 (WV2) satellite imagery in a semi-informal settlement. WV2 satellite imagery data was fused with gray-level co-occurrence matrices (GLCM) to enhance building extraction. The algorithms used include the Gaussian Mixture Model (GMM), Random Forest (RF), and Support Vector Machine (SVM). The results indicate that GLCM does not improve the detection of buildings when using the GMM algorithm, but it increases building detection with RF and SVM. The GMM algorithm achieved the highest average accuracy of 92% for building detection. However, SVM and RF have an overall accuracy of 79% and 70% respectively. Though RF did not perform very well in identifying individual buildings, its overall accuracy was high. The outcome indicates that machine learning algorithms can adequately map building footprints from high-resolution satellite imagery.

Item Type: Article
Keywords: Building detection, WorldView-2, machine learning, Gray-Level Co-Occurrence Matrix (GLCM)
Subjects: T Technology > TA Engineering (General). Civil engineering (General)
Divisions: Journal of Civil Engineering and Urbanism (JCEU)
Page Range: 149–157
Journal or Publication Title: Journal of Civil Engineering and Urbanism
Journal Index: Not Index
Volume: 14
Number: 3s
Publisher: Scienceline Publication
Identification Number: https://doi.org/10.54203/jceu.2024.14
ISSN: 2252-0430
Depositing User: Dr. Heydar Dehghanpour
URI: http://eprints.science-line.com/id/eprint/1314

Actions (login required)

View Item View Item