Journal
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
Volume 59, Issue 6, Pages 5206-5220Publisher
IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
DOI: 10.1109/TGRS.2020.3015878
Keywords
Spatial resolution; Remote sensing; Neural networks; Dictionaries; Transforms; Deep learning; multispectral (MS) image; panchromatic image; pansharpening; super-resolution (SR)
Categories
Funding
- Strategic Priority Research Program of the Chinese Academy of Sciences [XDA19090108]
Ask authors/readers for more resources
A novel pansharpening algorithm guided by a deep SR convolutional neural network is proposed in this study, which outperforms traditional and other deep-learning-based methods in improving spatial resolution of images. The algorithm includes components such as SR process, progressive pansharpening process, and high-pass residual module for enhancing spatial details from different types of satellite data.
Pansharpening and super-resolution (SR) methods share the same target to improve the spatial resolution of images. Based on this similarity, we propose and develop a novel pansharpening algorithm that is guided by a deep SR convolutional neural network. The proposed framework comprises three components: an SR process, a progressive pansharpening process, and a high-pass residual module. Specifically, the SR process extracts inner spatial detail that is present in multispectral images. Then, progressive pansharpening is used as a detailed pansharpening process, and the high-pass residual module helps by directly injecting spatial detail from panchromatic images. The performance of the proposed network has been compared with that of traditional and other deep-learning-based pansharpening algorithms based on QuickBird, WorldView-3, and Landsat-8 data, and the results demonstrate the superiority of our algorithm.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available