Learning Deep Pyramid-based Representations for Pansharpening

Deep learning-based pansharpening has emerged as a dynamic research area. Retaining spatial & spectral characteristics of panchromatic image and multispectral bands are a critical issue in pansharpening. This paper proposes a pyramid-based deep fusion framework that preserves spectral and spatial characteristics at different scales. The spectral information is preserved by passing the corresponding low-resolution multispectral image as residual component of the network at each scale. The spatial information is preserved by training the network at each scale with the high frequencies of panchromatic image alongside the corresponding low resolution multispectral image. The parameters of different networks are shared across the pyramid in order to add spatial details consistently across scales. The parameters are also shared across fusion layers within a network at a specific scale. Experiments show that the proposed architecture exhibits better performance than state-of-the-art pansharpening models. At reduced scale, the proposed scheme has enhanced the fusion quality in terms of universal quality index, spectral angle mapper, relative global error, and spatial correlation coefficient by \(9.6\%\), \(33.1\%\), \(36\%\), and \(11.2\%\), respectively. Similarly, at full scale, the fusion performance is improved in terms of spectral & spatial distortions, and no reference quality metrics by \(47.3\%\), \(36.7\%\), and \(9.5\%\), respectively.