Land Use Mapping of Sabzevar using Maximum Likelihood and Artificial Multilayer Perceptron Neural Network

Subject Areas :

Elahe Akbari ¹ , Majid Ebrahimi ² , Abolghasem AmirAhmadi ³

1 - MSc. Remote Sensing and GIS, Hakim Sabzevari University‚ Sabzevar‚ Iran.
2 - Ph.D. Student of Geomorphology, Hakim Sabzevari University‚ Sabzevar‚ Iran.
3 - Associate Professor‚ Geography Department‚ Hakim Sabzevari University‚Sabzevar‚ Iran

Received: 2013-02-18 Accepted : 2013-09-18 Published : 2014-02-20

Keywords: urban land use, satellite data, Maximum likelihood, Multilayer Perceptron Neural Network, overall accuracy,

Abstract :

Among the important factors in urban planning and management, particularly in line with the achievement of the sustainable development in the urban areas as well as regarding the optimal use of the land, is on-time access to the data of land cover conditions in these regions. The remote sensing data has a high potential for the preparation of the update urban land cover maps. In order to present on-time and digital satellite data, a variety of shapes and possibility of processing during land cover maps are of high significance. In order to use the satellite photos Landsat/ETM+ and two algorithm of supervised classification including the maximum likelihood and the artificial neural network, land cover maps were prepared. During classification, the neural network algorithm of a perceptron network with a hidden layer and 7 input neurons, nine middle neurons and 4 output neurons were used. The input neurons are the same in number as the bands of the Landsat photos and the number of output neurons are the same as land cover map classes. Eventually, land cover map of the region has been classified into four classes of residential areas, barren lands, plant coverage, and roads. In order to evaluate the correctness of the classification results, many photos have been taken using GPS. Using overall accuracy and Kappa Coefficient the precision evaluation results of these two methods indicate that perceptron neural network has an overall accuracy of 98/24 and Kappa Coefficient 97/03 compared to the algorithm of maximum likelihood with an overall accuracy of 94/23 and Kappa Coefficient 90 / 34 is of higher precision. The findings of this study also show that the classification method for multilayer perceptron neural network as compared with the maximum likelihood method is of higher separation and capability for preparing the land cover map in the urban regions.