Using mathematical algorithms for classification of LANDSAT 8 satellite images

Abstract

Satellite and aerial images are objective photographical representations of the reality from the field, related to spatial – temporal frames. The purpose of the present study is to create a comparative analysis of a LANDSAT 8 satellite image by supervised and unsupervised classification methods. The algorithms used in this research were maximum likelihood algorithm for supervised classification and ISODATA algorithm for unsupervised classification. From the multitude of possible combinations of spectral bands we chose the combination bands near infrared-red-green (543), image which has been subjected to the supervised and unsupervised classification process. This combination includes in addition the two bands from visible (R, G) the band near-infrared (NIR) which represents an area where the response of the vegetation is tinted and the cloudy formations are easier t penetrate. By analyzing the satellite image through supervised classification it enabled a result of 5 thematic classes defined by the user based on information in the field, leaving an unclassified fraction. Unsupervised Classification generated 16 classes covering the entire area of interest. Comparative analysis of satellite images by the two methods facilitates increased accuracy of image classification.