Land cover classification using apropriate fuzzy image classification tecniques: Aladağ case study

ABSTRACTMsc THESISLAND COVER CLASSIFICATION USING APROPRIATE FUZZYIMAGE CLASSIFICATION TECNIQUES: ALADAG CASE STUDYOnur ŞATIRDEPARTMENT OF LANDSCAPE ARCHITECTURE INSTITUTE OF NATURELAND APPLIED SCIENCES UNIVERSITY OF CUKUROVASupervisor: Assoc.Prof. Süha BERBEROĞLUYear: 2006 Page: 79Jury: Prof.Dr. M. Faruk ALTUNKASAAssoc.Prof. A. Oğuz DİNÇAssoc.Prof. Süha BERBEROĞLUThe aim of this study was to classify Envisat MERIS and Landsat ETM satellite imageryrecorded in August 2003 over the Aladağ in the upper basin of Seyhan River basin, usingfuzzy classification techniques such as, linear mixture modelling and artificial neuralnetworks. The images were classified successfully using these two fuzzy techniques. Fuzzyclassification techniques produced more accurate results than hard classification. LandsatETM imagery was classified using maximum likelihood classifier and the output wasresampled to 300 m to produce test data. Hard classification result was produced usingartificial neural network classifier and tested using 500 random points.Eight major land cover classes (including agriculture, bare ground, Pinus brutia, Pinusnigra, Cedrus libani, Abies sp., Juniperus sp., water) were classified within thisclassification. The classification accuracy of some land cover classes were below acceptablelevel as there was insufficient number of training pixels available. For this reason, overallaccuracy approach was applied to compare applicability potential of the techniques. Overallresults of soft classification from linear mixture modeling and artificial neural network andhard classification from artificial neural network were 82%, 81% and 57% respectively.It can be concluded that soft classifiers resulted more accurate classifications than hardclassifiers. Additionally, there is no significant difference between soft classification outputsfrom linear mixture modeling and artificial neural networks, however artificial neuralnetworks tackled pixels with high degree of mixing more accurately than linear mixturemodeling.Keywords: Fuzzy logic, image classification, linear mixture modeling, artificial neuralnetworks, remote sensing.II