Image classification involves extracting information from multiple-band satellite images. Statistical methods and machine learning algorithms have gained popularity for image classification in recent years. This study aims to perform image classification using statistical and machine learning algorithms and evaluate their performance. Specifically, three algorithms, Random Forest (RF), Support Vector Machines (SVM) and Multivariate Adaptive Regression Splines (MARS), were selected for this analysis. The study area covers the three districts of Samsun (Canik, İlkadım and Atakum) with the highest population density and various land cover types. Landsat-9 OLI-2 satellite image of September 30, 2022 was used in the analysis. Considering the CORINE classification categories, five classes were created, including water, forest, urban fabric, heterogeneous agricultural areas, and open spaces with little or no vegetation. The implementation of the three algorithms was carried out using RStudio software. The performance of these algorithms was evaluated using kappa statistics, a widely used measure of reliability in image classification. The MARS algorithm exhibited the highest level of success, achieving a kappa value of 0.847. The RF algorithm followed closely with a kappa value of 0.828. SVM ranked last in performance with a kappa value of 0.798. Consequently, the results indicate that the MARS algorithm outperforms the others. ORCID NO: 0000-0003-2625-3690
Anahtar Kelimeler: Image Classification, Classification Accuracy, Random Forest, Support Vector Machines, Multivariate Adaptive Regression Splines
|
