Comparison of neural network methods (fuzzy ARTMAP, Kohonen and Perceptron) and maximum likelihood efficiency in preparation of land use map

With the development of various methods in the field of satellite image classification and change detection, especially in recent decades, choosing the best and accurate method for preparation of lands use map and lands cover in different regions has become increasingly grown. For this purpose, classification of satellite images in Yazd plain-Ardakan was done as 7 categories of lands use including (poor pasture, bare lands, residential areas, sand dunes, Semi-dense pasture, agricultural and rocky lands) and then training samples were collected from the area by using 1: 20,000 aerial photos, satellite images, Google Earth and field visit. Then, by using the properties of the images, land-use classes in the study area was chose and after determining the classes` resolution, the classification in a supervised way and by using artificial neural network techniques, Fuzzy ARTMAP, Kohonen, Perceptron and maximum likelihood was done. The aim of this study was to compare the maximum likelihood algorithm and three methods of artificial neural network to classify the land cover in the study area (Yazd plain-Ardakan). The results of evaluation of the accuracy of these four methods showed that the overall accuracy of maximum likelihood classification methods, Kohonen, Fuzzy ARTMAP and Perceptron neural network method is 89/0, 89/0, 87/0 and 23/0 respectively and their kappa coefficient is 91%, 80%, 73% and 21% respectively. Therefore, according to the results, the maximum likelihood method has the best performance in land use mapping in the study area.