Hassan Fathizad; Ata Safari; masoud Bazgir; Gholamreza Khosravi
Volume 23, Issue 4 , March 2017, , Pages 729-743
Abstract
Image classification is always one of the most important issues in remote sensing, and the obtained information from image classification is widely used in this field and other applications like urban planning, natural resource management, agriculture, etc. Since the main purpose of processing satellite ...
Read More
Image classification is always one of the most important issues in remote sensing, and the obtained information from image classification is widely used in this field and other applications like urban planning, natural resource management, agriculture, etc. Since the main purpose of processing satellite images is preparing subjective and practical maps, choosing a suitable classification algorithm has an essential role. This paper studies the efficacy of Support Vector Machines (SVM) algorithm regarding satellite image classifications and compares it to artificial neural network algorithm. SVM is a group of classified and observed mechanical learning algorithms, used in remote sensing. In this study, SVM algorithms were employed for land use classification of Meymeh area using ETM+ landsat data. The classification via SVM was automatically performed by three types of linear Kernel, polynomial, and radial basis. Besides, the performance of this method was compared to that of artificial neural network classification method. Results showed that the average overall accuracy and Kappa coefficient of SVM algorithms, including linear Kernel, polynomial and radial basis, were respectively 9 percent and 12 percent more efficient than artificial neural network classification. Consequently, this study substantiates the efficiency and sufficiency of SVM algorithms in classification of remote sensing images.