Evaluating the Efficiency of Four Artificial Neural Network Methods in Preparing Land Cover/Land Use Map Using ETM+ Data Case study: Doiraje, Mehran and Sarableh

Abstract

Land use/cover maps resulting of satellite images play an important role in assessing the land use/ land cover at regional and national levels. Over the last years, many applications of neural network classifiers for land use classification have been reported in the literature, but afew studies have assessed their comparison. In this study, firstly, geometric correction was performed on ETM+ data. Then, with field surveyings, the various land cover classes were defined and training areas were selected. The main Objective of this study is to compare four artificial neural network methods for land cover classification in Doiraj, Mehran and Sarableh region of Ilam province with various climatic conditions. In this study, we have used four artificial neural networks methods of Fuzzy Artmap, multi-layer perceptron, Kohonen and radial basis function. The results obtained of accuracy assessment of classified images showed that fuzzy Artmap classification algorithm with the overall accuracy 94.84 and kappa coefficient 0.93% have the highest accuracy than other methods. Accuracy overall difference in this approach than multi-layer percepteron method was 11.44 and Kappa coefficient 0.18, Compared to kohonen's 17.30 and 0.23% and rather than radial basis function 31.01 and 0.36%, respectively. In this study, the highest accuracy was related to fuzzy Artmap artificial neural network. Therefore, this study proves the efficiency and capability of fuzzy Artmap neural network algorithm in classification of remote sensing images.

Keywords

• land use
• image classification
• fuzzy artmap classifier
• radial basic function classifier
• kohonen classifie
• perceptron multi-layer classifier