Analysis of Image Classification Methods for Remote Sensing

• Published in: Experimental techniques (Westport, Conn.) Vol. 36; no. 1; pp. 18 - 25
• Main Authors: Ayhan, E., Kansu, O.
• Format: Journal Article Magazine Article
• Language: English
• Published: Oxford, UK Blackwell Publishing Ltd 01.01.2012; Springer International Publishing; Springer Nature B.V
• Subjects: Algorithms; Artificial Neural Networks; Characterization and Evaluation of Materials; Chemistry and Materials Science; Classification; Fuzzy Logic; Image classification; Landsat satellites; Learning theory; Materials Science; Maximum Likelihood Classification; Networks; Neural networks; Remote Sensing; Technical Article; Remote Sensing; Artificial Neural Networks; Fuzzy Logic; Maximum Likelihood Classification; Classification
• ISSN: 0732-8818; 1747-1567
• DOI: 10.1111/j.1747-1567.2011.00719.x

Maps of land usage are usually produced through image classification that is a process on remotely sensed images for preparing the thematic maps. In this study, multispectral IKONOS II and Landsat imagery data were classified with the methods of artificial neural networks, standard maximum likelihood classifier, and fuzzy logic method. While back-propagating learning algorithm was used for artificial neural network method, Sugeno-type fuzzy model was used for the application of fuzzy logic method. Also, the determination of the optimum design of ANN classification was aimed by using ANN learning algorithms and designating different networks. Comparisons were made in terms of classification accuracy that is the validation tool for the process of image classification. Results show that artificial neural network classification is more robust than the standard maximum likelihood method and fuzzy logic method. However, determining the optimum network structure is a cumbersome but necessary stage in the classification of ANN.