Assessment of ASTER Thermal Bands Capabilities in Enhancing Accuracy of Support Vector Machine and Maximum Likelihood Classification Algorithms in Arid Regions

Data recorded in the thermal band satellites, are commonly used in the thermal characteristics of different phenomena, especially in desert areas. Classification is one of the most important steps in the use of satellite imagery. In this study, using ASTER data TERRA satellite on 22 August 2001, East and North East Kashan Salt Lake, were studied to find the role of quantity bands of this sensor to increase the resolution accuracy of phenomena in the maximum likelihood (MLK) and support vector machine (SVM). For this purpose, after the initial radiometric and geometric corrections, eight approaches were chosen with different band combinations step by step in order to investigate the quantitative role of each band to increase the classification accuracy and then the classification accuracy using Kappa index, user and producer accuracy were evaluated. The classification algorithm results showed that the support vector machine algorithm rather than maximum likelihood algorithm has slightly better results. Generally, the use of all spectral and thermal bands (14 bands) had a highest accuracy for both Kappa support vector machine algorithm (83.04) and maximum likelihood (90.82). While the Kappa accuracy by 15% in both algorithm with elimination of all thermal bands (bands 10, 11, 13.12 and 14) was reduced. The maximum likelihood algorithm had the greatest impact on increasing the accuracy of Kappa index bands of thermal 14 (between 8 to 10 percent) and in the support vector machine algorithm bands of 10 and 14 (7%). Finally, it was proven that ASTER sensor due to wide spectrum range in the wavelengths of thermal infrared has very high potential in increasing accuracy coefficient of classification Kappa.