ABSTRACT With the recent advances made by deep neural networks for image processing applications, researchers have begun exploring this avenue for pan-sharpening and obtained remarkable results. However, the existing methods are generally limited by their shallow network architectures, resulting in insufficient reconstruction capacities. To produce high-quality pan-sharpened images, this paper proposes a novel neural network for pan-sharpening that uses a multilevel shortcut link and dense block to improve the depth of the neural network. Notably, this paper designs two shortcut links to compensate for the loss of spectral and spatial features during the training process. The experimental results show that the proposed method is superior to the state-of-the-art methods in both subjective visual and objective assessments.

中文翻译：

---

用于全色锐化的多级密集神经网络

摘要 随着深度神经网络在图像处理应用方面的最新进展，研究人员已经开始探索这种全色锐化的途径并取得了显着的成果。然而，现有方法普遍受到其浅层网络架构的限制，导致重建能力不足。为了产生高质量的全色锐化图像，本文提出了一种新的全色锐化神经网络，它使用多级快捷链接和密集块来提高神经网络的深度。值得注意的是，本文设计了两个快捷链接来补偿训练过程中光谱和空间特征的损失。实验结果表明，所提出的方法在主观视觉和客观评估方面均优于最先进的方法。