Pan-Sharpening Using Fusion of 2D-DFRFT, 2D-DWT, and CNN

In recent years, the demand for high-resolution remote sensing imaging has increased in a variety of fields, including environmental monitoring, urban planning, agriculture, and disaster management. Multispectral imaging, which gathers information at several wavelengths, can provide useful insights for such applications. However, the spatial resolution of multi- spectral images frequently falls short of the necessary criteria for thorough analysis. Pansharpening approaches overcome this issue by combining high-resolution panchromatic (PAN) images with lowerresolution multispectral data, thereby improving spa- tial details while keeping spectral features. In this article, we present a unique pan-sharpening technique that makes use of the synergies between the two-dimensional discrete fractional Fourier transform (2D-DFRFT), the twodimensional discrete wavelet transform (2D-DWT), and convolutional neural network (CNN) models. By combining these complex mathematical transforms with deep learning approaches, we provide a holistic solution for improving both spectral and spatial resolution. Our method's main innovation is its multi-stage procedure, which starts with converting the PAN and multispectral images using 2DDFRFT and 2D-DWT, then extracts features using a CNN model. Subse- quent high-pass filtering of the converted images refines spatial information, allowing for a more precise fusion process. The inclusion of angle parameters associated with both 2D-DFRFT and 2D-DWT provides an additional degree of freedom, allowing for finer control over the fusion process and producing better results. We test the performance of our proposed pan-sharpening system using real-world satellite pictures from IKONOS and GeoEye-1 via extensive experiments in MATLAB.