عنوان مقاله [English]
نویسنده [English]چکیده [English]
In order to classify satellite image of the ETM+, classes of land uses were classified in six groups as agricultural lands, rangeland, forest, land barren, garden, lake and then training samples were collected from study area using Google Earth satellite image and the field visits. At the next stage, by using the images characteristics, According to results, the tree classification with three splitting methods(gain ratio, entropy, and gini) produced the overall accuracy of 87 and kappa coefficient 0.84, respectively, while, methods of fuzzy Artmap and maximum likelihood were classified with overall accuracies of 84, 81 and Kappa coefficients of 0.81, 0.78, respectively. Thus, the splitting methods of tree classification (average overall accuracy of 3% and Kappa coefficient of 3% in comparing with to methods of fuzzy Artmap artificial neural network, and average overall accuracy 6% and kappa coefficient of 6%) than likelihood maximum classification for the data series used in this study were of higher accuracy. The efficiency of the tree classification with gain ratio splitting to be roughly comparable to the fuzzy Artmap ANN method and this reflects the high efficiency of fuzzy Artmap neural network. Finally, we can say that among three splitting methods used in this study, the gini splitting method with overall accuracies of 6% and 2% and kappa coefficients of 7% and 2% higher than the entropy method respectively has better performance. This study shows tree classification methods have many advantages over the other methods of classification such as fuzzy Artmap artificial neural network and maximum likelihood methods. They are computationally fast (Unlike artificial neural network methods) and make no statistical assumptions regarding the distribution of data (Unlike the maximum likelihood method). As the result, it can be said that, tree classification is a good alternative for other methods of classification.