Research on Worldview-3 Panchromatic and Shortwave Infrared Image Fusion Method

Abstract

Worldview 3 is one of the most advanced high-resolution optical satellites. Aiming at the problems of large difference in spatial resolution between panchromatic band and short wave infrared (SWIR) band of Worldview 3 remote sensing satellite data and inconsistent spectral range, resulting in massive effect of fusion results and limited effect of spatial resolution enhancement, pannet network training is used for fusion. Firstly, the network reduces the spatial resolution of panchromatic band and realizes the preliminary integration with SWIR band; Then the preliminary fusion results are fused with the original resolution panchromatic band again. For spectral preservation, pannet adds the sampled multispectral image to the network output, which propagates the spectral information directly to the reconstructed image. The network trains the network parameters in the high pass filter domain rather than the image domain, so as to preserve the spatial structure. The results show that deep learning can achieve good results in image fusion. Pannet network structure can effectively enhance the spatial resolution of SWIR band, and also has a certain reference significance for the integration of traditional panchromatic and short wave infrared band.