This paper proposes a classification method that fuses superpixels-segmentation information from an RGB image with a hyperspectral image without estimating the high spatial-spectral resolution cube. This methodology improves the classification accuracy while boosting the performance.