Application of machine learning methods in the classification of satellite images

Abstract

Machine learning, as a specific domain within artificial intelligence, opens new horizons for both theoretical and experimental research in remote sensing, particularly in satellite imagery classification. This study focuses on applying machine learning methods, specifically decision trees and support vector machines, to classify satellite images. The analysis uses the SAGA GIS software on LANDSAT 8 OLI Level 2A satellite images. Satellite image classification encompasses two primary groups of computer operations: unsupervised (automatic or formal) and supervised (semi-automatic or logical) classification. This research executes the practical classification of satellite images by applying the aforementioned machine learning methods. The results indicate that the obtained classified rasters not only align with but also fully replace existing classification and identification methods of geospatial objects. Consequently, this research contributes to a significant advancement in collecting and analysing geospatial data.