Unmixing-based fusion of hyperspectral images for classification

Hyperspectral images compared to standard multispectral images can resolve materials on earth with higher accuracy. However, due to low spectral resolution of hyperspectral images, there exists the problem of `mixed pixels´. Fusion of high spectral resolution images with high spatial resolution images is known to yield rich information content products. In this study we propose unmixing-based fusion of hyperspectral images with multispectral images to improve classification accuracy. Accordingly; classification of fused images yielded results with higher spatial detail compared to that of low resolution hyperspectral image. Unmixing-based fused images proved to improve the classification accuracy compared to classification results obtained from conventionally fused images. End member selection approach is observed as another factor influencing classification accuracy.