Pan-Sharpening Based on a Deep Pyramid Network

The purpose of remote-sensing image fusion(ie, pan-sharpening) is to generate high-resolution multispectral images using high-resolution single-band panchromatic(PAN) images and low-resolution multispectral images(MS). In this paper, a pan-sharpening algorithm based on deep pyramid network is proposed, which reconstructs high-resolution multispectral images through layer-by-layer upsampling. To preserve detail information, aiming at the problem that the panchromatic image and the multispectral image differ too large in scale, this paper uses the deep pyramid network to fuse the details of the panchromatic image in a multi-scale fusion strategy. For spectral preservation, this paper uses the deconvolution layer instead of the traditional super-resolution algorithm to upsample low-resolution multispectral images. Finally, the two parts are added to obtain the fused image. The experimental results on GeoEye-1 datasets demonstrate that the proposed algorithm achieve competitive performance in comparison with BDSD, PRACS, PNN and PanNet.