This paper presents a new pansharpening method based on convolutional neural networks (CNNs). The authors adapt a three-layer architecture originally proposed for super-resolution tasks to the pansharpening problem. To enhance performance without increasing complexity, they augment the input with several nonlinear radiometric indices commonly used in remote sensing. Experiments on three datasets—Ikonos, GeoEye-1, and WorldView-2—show that the proposed method outperforms existing state-of-the-art techniques in both full-reference and no-reference metrics, as well as visually. The method leverages domain-specific knowledge to improve the network's ability to capture relevant features, such as vegetation and water signatures, from the multispectral components. The results demonstrate the effectiveness of using deep learning for pansharpening tasks, particularly when combined with domain-specific features.This paper presents a new pansharpening method based on convolutional neural networks (CNNs). The authors adapt a three-layer architecture originally proposed for super-resolution tasks to the pansharpening problem. To enhance performance without increasing complexity, they augment the input with several nonlinear radiometric indices commonly used in remote sensing. Experiments on three datasets—Ikonos, GeoEye-1, and WorldView-2—show that the proposed method outperforms existing state-of-the-art techniques in both full-reference and no-reference metrics, as well as visually. The method leverages domain-specific knowledge to improve the network's ability to capture relevant features, such as vegetation and water signatures, from the multispectral components. The results demonstrate the effectiveness of using deep learning for pansharpening tasks, particularly when combined with domain-specific features.