Farklı Piksel Tabanlı Sınıflandırma Yöntemlerinin Arazi Kullanımı ve Arazi Örtüsü Belirlemedeki Performansının İncelenmesi

Abstract

With the development of photogrammetry and remote sensing techniques, data collection has become easier. However, due to the large size of the data collected, extracting meaningful data from the data set has become a popular topic. Nowadays, the development of digital image processing techniques has contributed to the determination of land cover land use (LCLU) through digital images. In this study, a supervised classification was made over the orthophoto view to distinguish different land object classes in a campus area. The purpose of the study is to examine the performance of the three popular supervised classification techniques that are maximum likelihood, minimum distance, and mahalanobis distance methods. In the study, a confusion matrix was produced, and overall accuracy and overall kappa were calculated with manually generated ground truth data. According to results, the highest overall accuracy was calculated for maximum likelihood classification with a rate of 84.5 % and the minimum distance method has the lowest overall accuracy (43%). The research denotes that due to the lack of spectral information the supervised classification methods generate omission and commission errors. This fact has a direct effect on overall accuracy calculation.