Fast Full-Resolution Target-Adaptive CNN-Based Pansharpening Framework

In the last few years, there has been a renewed interest in data fusion techniques, and, particular, pansharpening due to paradigm shift from model-based data-driven approaches, supported by recent advances deep learning. Although plethora of convolutional neural networks (CNN) for have devised, some fundamental issues still wait answers. Among these, cross-scale and cross-datasets generalization capabilities are probably most urgent ones since current trained at different scale (reduced-resolution), general, they well-fitted on datasets but fail others. A attempt address both these leverages target-adaptive inference scheme operating with suitable full-resolution loss. On downside, such an approach pays additional computational overhead adaptation phase. this work, we propose variant method effective target-adaptation that allows reduction time factor ten, average, without accuracy wide set experiments carried out three datasets, GeoEye-1, WorldView-2 WorldView-3, prove gain obtained while keeping top scores compared state-of-the-art methods, deep-learning ones. The generality proposed solution also validated, applying new framework CNN models.