Mohsen Farzin
Abstract
The aim of this study was to investigate the ability of different supervised and unsupervised classification algorithms of remote sensing data for detecting and separating of land cover on Beshar River Basin using Landsat 8 data. For this purpose, after checking the geometric accuracy and radiometric-atmospheric ...
Read More
The aim of this study was to investigate the ability of different supervised and unsupervised classification algorithms of remote sensing data for detecting and separating of land cover on Beshar River Basin using Landsat 8 data. For this purpose, after checking the geometric accuracy and radiometric-atmospheric corrections on satellite data, the data set was created to the combination of spectral bands (bands 2, 3, 4, 5, 6, 7 and 8) and thermal (band 10). Next, pixel-based classification using supervised algorithms including maximum likelihood, support vector machine, mahalanobis distance, minimum distance, neural network, parallelepiped, spectral angle mapping, spectral information divergence, binary coding, and unsupervised algorithms including K-Means and IsoData was done. The accuracy of the algorithms for identifying each land use /land cover based on the error matrix analysis was evaluated using the producer's accuracy, user accuracy and overall accuracy based on the omission and commission errors, and the kappa coefficient. The results showed that the most appropriate algorithm for separation and identification of land use/land cover including agriculture, construction, cliff, forest, orchard, rangeland, water body and fallow is maximum likelihood, mahalanobis distance, maximum likelihood, mahalanobis distance, neural network, support vector machine, support vector machine, and maximum likelihood, respectively. The percentage of overall accuracy and Kappa coefficient shows that the four algorithms including maximum likelihood, support vector machine, mahalanobis distance and neural network with overall accuracy 77.25, 75.9, 69.59, 68.26 and the Kappa coefficient 0.72, 0.69, 0.63, 0.58, respectively, is better than other algorithms. Generally, the integration of appropriate classification algorithms in mountainous areas increases classification accuracy and will have better results.
