Abstract. Most existing multispectral fusion algorithms often suffer from spectral or spatial information distortion. Driven by this motivation, we propose an edge-guided multispectral (MS) image fusion algorithm. In particular, it combines the advantages of generative adversarial networks and improved fusion frameworks, so the merged image can better preserve the spectral information of the original multispectral image while injecting spatial detail information. Specifically, first, an MS image with more image detail is generated using the generated confrontation network for preliminary reconstruction. The panchromatic image edge information and the antagonistic learning strategy are introduced for the robust multispectral image reconstruction. Then, using the reconstructed MS image and the general component substitution image fusion framework, the whole fusion system of this paper is constructed. An enhancement operator is introduced to inject spatial details. Our extensive dataset evaluations show that our approach performs better in terms of high objective quality and human visual perception than several of the most advanced fusion methods.

中文翻译：

---

边缘引导的多光谱图像融合算法

摘要。大多数现有的多光谱融合算法经常受到光谱或空间信息失真的影响。在此动机的推动下，我们提出了一种边缘引导的多光谱 (MS) 图像融合算法。特别是它结合了生成对抗网络和改进融合框架的优点，因此合并后的图像在注入空间细节信息的同时，可以更好地保留原始多光谱图像的光谱信息。具体来说，首先使用生成的对抗网络生成具有更多图像细节的 MS 图像进行初步重建。引入了全色图像边缘信息和对抗学习策略，用于鲁棒的多光谱图像重建。然后，使用重建的 MS 图像和通用分量替换图像融合框架，构建了本文的整个融合系统。引入了增强算子来注入空间细节。我们广泛的数据集评估表明，与几种最先进的融合方法相比，我们的方法在高客观质量和人类视觉感知方面表现更好。