Impact Assessment of image fusion methods on the final accuracy of remote sensing image classification with medium resolution (case study: Tehran region)

Today, with the technological advances in earth monitoring and observation technologies, satellite images have been utilized in different remote sensing applications. These satellite images can acquire panchromatic and multispectral images, which have different spatial and spectral resolution. Panchromatic images have high spatial and low spectral resolution, and multispectral images have high spectral and low spatial resolution. For obtaining an image with high spatial and spectral resolution at the same time, pan-sharpening techniques can be used. Pan-sharpening procedure is done in the preprocessing stage of digital image processing in the applications like generating land cover and land use maps. These techniques provide better visual interpretations, which is vital for land cover and land use extraction. After obtaining the pan-sharped image, the land cover and land use maps can be created using classification techniques. In this study, a Landsat ۹ image, which is a medium resolution image, was used to obtain pan-sharped image of Tehran region. The pan-sharpening methods of IHS, Brovey, Gram-Schmid, Nearest Neighbor Diffuse and CN-Spectral were used to fuse panchromatic and multispectral images. The pan-sharped images were assessed by ERGAS, UIQI, Correlation Coefficient (CC), BIAS and RASE indexes to determine the optimum pan-sharpening method. The results show that the Gram-Schmid technique had better results in those indexes than other pan-sharpening methods. This Gram-Schmid pan-sharped image was then classified by Maximum Likelihood and Support Vector Machine (SVM) supervised to create land cover and land use map of Tehran area. These classified maps were assessed by confusion matrix, overall accuracy and Kappa coefficient to determine which classification method had better accuracy in creating land cover and land use maps. The results showed that Maximum Likelihood method was more accurate than Support Vector Machine (SVM) to obtain land cover and land use information from the Gram-Schmid pan-sharped image.