Determining the optimal method for classification and mapping of land use/land cover through comparison of artificial neural network and support vector machine algorithms using satellite data (Case study: International Hamoun wetland)

Images classification is one of the important techniques for interpretation of satellite images that is widely used in survey of earth changes. In the meantime, satellite data has been recognized as the best tool for detection and evaluation of changes due to its update information, low costs and variety of forms. Therefore, land use/land cover map is one of the most important information required by the environmental managers and planners. On the other hand, in recent years, artificial neural network method has been used widely for the classification of satellite data. The aim of this study is to compare three different methods for land cover classification using 2014 OLI image over a 26-year period. In this study, digital data of OLI (2014) sensor was used in order to optimize image classification method. Initially, the image was corrected in terms of geometry and radiometry in the ENVI software. Then IDRISI software was used for image classification using three different methods fuzzy artmap, multilayer perceptron artificial neural networks and support vector machine. Finally, land cover maps were classified into five categories: water, vegetation, canebrake, barren lands and saline lands. To evaluate accuracy with the help of user accuracy, producer accuracy, overall accuracy, kappa coefficient and error matrix, the created map was compared with the ground reality map created by GPS, Google Earth images and field observations. The results of image accuracy evaluation showed that among the applied methods the fuzzy artmap algorithm had the highest accuracy in classification of satellite data with an overall accuracy of 94.68 and kappa coefficient of 0.91 compared to both multilayer perceptron artificial algorithm with an overall accuracy of 92.99 and kappa coefficient of 0.89 and support vector machine with an overall accuracy of 90.93 and kappa coefficient of 0.85. This study showed that classification of fuzzy artmap artificial neural network algorithm has a high capability to create the land cover map with high accuracy.