Abstract:
Pansharpening process aims at fusing low-spatial/high-spectral resolutions multispectral/hyperspectral (MS/HS) remote sensing data with high-spatial resolution and withou...Show MoreMetadata
Abstract:
Pansharpening process aims at fusing low-spatial/high-spectral resolutions multispectral/hyperspectral (MS/HS) remote sensing data with high-spatial resolution and without spectral diversity panchromatic (PAN) ones. This paper explores different data preparation possibilities, learning strategies and architectures, used in the convolutional neural network (CNN) approaches, for improving the performance of the pansharpening process of remote sensing MS/HS data. Also, in this paper, the super resolution CNN (SRCNN) architecture is adapted by adding a normalization step in the training phase of the CNN-based pansharpening process. Then, training datasets are prepared for fitting the generalization need. Experiments based on multi-source datasets are performed to evaluate the performance of the proposed SRCNN-based pansharpening architecture. The preliminary results are promising since they show that the proposed approach is competitive with some literature methods.
Published in: 2018 9th Workshop on Hyperspectral Image and Signal Processing: Evolution in Remote Sensing (WHISPERS)
Date of Conference: 23-26 September 2018
Date Added to IEEE Xplore: 27 June 2019
ISBN Information: