Volume 11, Issue 1 (Spring & Summer 2025)
KJES 2025, 11(1): 201-219 |
Back to browse issues page
Download citation:
BibTeX | RIS | EndNote | Medlars | ProCite | Reference Manager | RefWorks
Send citation to:
BibTeX | RIS | EndNote | Medlars | ProCite | Reference Manager | RefWorks
Send citation to:
Talkhablou M, Farmahinifarahani M, Siahmansouri S. A python-based framework for land cover classification in engineering geology: A comparative assessment of SVM, K-means, and spectral indices. KJES 2025; 11 (1) :201-219
URL: http://gnf.khu.ac.ir/article-1-2951-en.html
URL: http://gnf.khu.ac.ir/article-1-2951-en.html
1- Kharazmi University , Talkhablu@gmail.com
2- Kharazmi University
2- Kharazmi University
Abstract: (451 Views)
Accurate land cover classification is a fundamental step in engineering geology studies, particularly for assessing slope instability and mass movements. With the growing availability of satellite data and machine learning tools, automated and reproducible classification frameworks have become essential. This study presents a comprehensive Python-based framework for land cover classification, comparing the performance of two machine learning algorithms, Support Vector Machine (SVM, supervised) and K-means clustering (unsupervised), against traditional spectral indices (NDVI, NDWI, UI, SAVI) using Landsat 8 imagery. The study area is located in East Azerbaijan Province, Iran, covering approximately 80×70 km with diverse land cover types, including vegetation, bare soil, urban areas, and surface water. Prior to classification, data underwent several preprocessing steps: gamma correction for visual enhancement, Min-Max normalization for data scaling, and Principal Component Analysis (PCA) for dimensionality reduction and multicollinearity mitigation. PCA retained components explaining at least 95% of total variance. Classification was performed on four main classes. Results were evaluated using Overall Accuracy (OA), Kappa Coefficient, and weighted Precision, Recall, and F1-Score. The SVM algorithm, using an RBF kernel, achieved the highest accuracy with 84% OA and a Kappa of 0.81, demonstrating superior ability in defining clear class boundaries, particularly in distinguishing urban areas from bare soil. In contrast, K-means clustering yielded 73% OA and a Kappa of 0.68, with noticeable class overlap. Spectral indices alone provided a baseline accuracy of ~65%, but their integration with machine learning models significantly improved performance. The findings confirm that supervised machine learning models, particularly SVM, outperform unsupervised clustering and standalone spectral indices. However, K-means remains viable in data-scarce scenarios. The proposed Python-based workflow offers a reproducible, transparent, and efficient approach for land cover analysis, making it a valuable tool for engineering geology applications such as landslide susceptibility mapping.
Keywords: Mass movements, K-means clustering, Support Vector Machine (SVM), spectral indices, machine learning, PCA, Python, land cover classification
Type of Study: Original Research |
Subject:
Engineering Geology
Received: 2025/07/12 | Accepted: 2025/09/9 | Published: 2025/09/21
Received: 2025/07/12 | Accepted: 2025/09/9 | Published: 2025/09/21
Send email to the article author
| Rights and permissions | |
 |
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License. |
