Spectral versus Spatial Pansharpening Methods for Classification Accuracy Enhancement

Abstract

Generally, the fusion process introduces two main factors into the resampled image, namely, the spectral distortion and the spatial enhancement. In this paper, we investigate the use of the pansharpening process to enhance the classification accuracy. Therefore, five pansharpening methods were used, namely, the PCA and IHS,from spectral methods and the HPF, HFM and ATW from the spatial methods. For the classification, two major classifiers were used, the SVM and the MLP. The classification was evaluated on two test sets, the first one lies inside the features while the second one is designed on the edges of features, to test both, the spectral and the spatial qualities. It was found that, for the first test set, the resampled image presents the best results followed by the HFM method. For the second one, the PCA method provides the best results. The ATW method is the poorest one in both cases. On the other hand, the evolution of the classification accuracy as a function of the spectral and spatial qualities, reveals no correlation to help in the prediction of the classification quality from the fusion results. Finally, we recommend the use of the PCA method for the classification of fused images.