MIRVJEN ULQINAKU, ANA KTONA
KEYWORDS : Raster classification, Maximum Likelihood, K-Nearest Neighbors, Random Trees, Support Vector Machine, Entropy
ABSTRACT :
Classification of ortho-imagery is a fundamental task in remote sensing applications, aiding in land cover analysis, urban planning, and environmental monitoring. In this article, we evaluate the performance of various classification algorithms in ArcGIS for five distinct land cover classes, using both Visible and InfraRed bands. The evaluated algorithms include Maximum Likelihood, K-Nearest Neighbors, Random Trees, and Support Vector Machine. In addition, this article will further focus on the performance of the Random Trees algorithm, particularly examining its utilization of entropy. Ortho-imagery datasets were acquired and preprocessed to ensure consistency and accuracy across all tests. Each classification algorithm was trained and tested using the same dataset and performance metrics such as overall accuracy, kappa coefficient, precision, recall and F1-score. Results reveal notable variations in classification accuracy among the algorithms tested, on both bands. For the visible band, Random Trees algorithm emerged with the best performance, followed by the Support Vector Machine algorithm. In contrast, for the infrared band, the Support Vector Machine algorithm exhibited the highest accuracy among the classifiers studied, closely followed by Maximum Likelihood algorithm. These findings provide valuable insights in remote sensing applications, aiding in the selection of appropriate classification algorithms for ortho-imagery analysis across multiple spectral bands.