Turkish Journal of Remote Sensing and GIS, vol.1, no.1, pp.34-44, 2020 (Peer-Reviewed Journal)
Image fusion is one of the most common techniques used to enhance the interpretability and functionality of remotely sensed data. The aim of this study was to improve the performance of the SVM (Support Vector Machines) classifier with the aid of texture features (TF) extracted from fused images. As a first step, the spatial resolution of the WorldView-2 MS (multispectral) imagery was increased by fusing it with a WorldView-2 PAN (panchromatic) image using the PCA (Principal Component Analysis), WSB (Wavelet Single Band), GS (Gram-Schmidt), BRV (Brovey), EHL (Ehlers), HCS (Hyperspherical Colour Space), HPF (High-Pass Filtering) and MCV (Multiplicative) algorithms. A PCA transform was then applied on all fused images. The first principal component obtained from each fused image was used to extract the Gabor TFs. As a final step, extracted Gabor TFs were combined with the original MS image. Resultant images were classified with the SVM algorithm to investigate to what degree the used methodology affect the classification accuracy. The results showed that the GS fusion-based Gabor TFs provided the greatest classification accuracy increase with 19.3%, whereas the PCA fusion-based Gabor TFs resulted in the second best classification accuracy increase with 18.7%.