Deep Learning Based Image Fusion Applied on Landsat-8 Archival Data Using UVCGAN

Image fusion is required to maximise the utilisation of the different satellite sensors and enhance the usage and quality of the images obtained from them. One can get over obstacles like varied geographical resolutions, spectral bands, and temporal coverage by fusing images from multiple satellites, such as Landsat 8 and Sentinel 2. For various applications that involve observation of the Earth like land cover classification, monitoring of vegetation, disaster monitoring, and urban planning, the fusion process aims to produce composite images that integrate the spectral richness, spatial detail, and temporal frequency of the input images. The creation of Sentinel-like imagery from Landsat8 data enhances its potential for Earth observation applications, making the problem of Landsat 8 and Sentinel 2 imagery fusion very important to solve. A deep learning methodology is required to solve this problem because of its capacity to discover complex correlations and patterns within the data. In this paper, a U-Net Vision Transformer Cycle-consistent Generative Adversarial Network (UVCGAN) - based image fusion method that effectively transforms Landsat 8 images into Sentinel-like imagery is pro- posed. The results obtained from the experiments illustrate the efficiency and effectiveness of the proposed method in generating high-quality composite images with improved spectral and spatial characteristics.