Evaluating non-parametric supervised classification algorithms in land cover map using LandSat-8 Images

The aim of this study was to evaluate the efficiency of three support vector machine algorithms, fuzzy decision trees and neural networks for mapping land vegetation map of Arakvaz watershed using OLI sensor of Landsat images (2014). Geometric correction and image pre-processing were utilized to determine the training samples of land vegetation classes for the classification operations. Sample resolution in the vegetation classes has been evaluated using a statistical divergence index. On the next stage, to evaluate the accuracy of algorithm's classification results, ground truth map with the dimensions of 550 m was designed using systematic approach and land vegetation types in the sampling plots were determined. Finally, the efficiency of each classification methodwas investigated bysuch criteria as overall accuracy, kappa coefficient, producer accuracy and user accuracy.Comparing the accuracy and kappa coefficient obtained for three categories with a proper band set in comparison with the ground truth map indicates that the Support Vector Machine (SVM) classifier with overall accuracy of 91.26% and kappa coefficient of 0.8731 has had more appropriate results than other algorithms. The results showed that the separation and classification of forest landswith high accuracy have beenperformedas compared to the other land use classes.