A new pansharpening method is proposed based on convolutional neural networks (CNN). The method adapts a three-layer architecture originally designed for super-resolution to the pansharpening task. To enhance performance without increasing complexity, the input is augmented with nonlinear radiometric indices typical of remote sensing. Experiments on three datasets show the method's effectiveness, outperforming existing techniques in both full-reference and no-reference metrics, as well as visually. The method uses a CNN-based architecture, with modifications to incorporate domain-specific knowledge from remote sensing. The input includes multispectral and panchromatic images, and the CNN processes them to produce high-resolution multispectral images. The method is tested on data from Ikonos, GeoEye-1, and WorldView-2 sensors, achieving significant improvements in performance. The CNN is trained using the Wald protocol, which involves downscaling and interpolating images to create a reference for training. The method is evaluated using various metrics, including full-reference and no-reference measures, and visual inspection confirms its effectiveness. The results show that the proposed method provides high-quality pansharpened images with minimal distortion and improved spatial and spectral resolution. The method is implemented using a CNN architecture, with training on GPU and testing on CPU. The results demonstrate the effectiveness of the proposed method in pansharpening, with significant improvements over existing techniques.A new pansharpening method is proposed based on convolutional neural networks (CNN). The method adapts a three-layer architecture originally designed for super-resolution to the pansharpening task. To enhance performance without increasing complexity, the input is augmented with nonlinear radiometric indices typical of remote sensing. Experiments on three datasets show the method's effectiveness, outperforming existing techniques in both full-reference and no-reference metrics, as well as visually. The method uses a CNN-based architecture, with modifications to incorporate domain-specific knowledge from remote sensing. The input includes multispectral and panchromatic images, and the CNN processes them to produce high-resolution multispectral images. The method is tested on data from Ikonos, GeoEye-1, and WorldView-2 sensors, achieving significant improvements in performance. The CNN is trained using the Wald protocol, which involves downscaling and interpolating images to create a reference for training. The method is evaluated using various metrics, including full-reference and no-reference measures, and visual inspection confirms its effectiveness. The results show that the proposed method provides high-quality pansharpened images with minimal distortion and improved spatial and spectral resolution. The method is implemented using a CNN architecture, with training on GPU and testing on CPU. The results demonstrate the effectiveness of the proposed method in pansharpening, with significant improvements over existing techniques.