Fourier-Transform-Based Unmixing Method for Fusion of Multiresolution Satellite Images

Many applications, such as agriculture and water monitoring, require frequent observations, as well as high spatial resolution. In practice, it is difficult to achieve both high resolution and temporal coverage for satellite sensors. Accordingly, several fusion algorithms for spatial and temporal images have been proposed. Among them, the unmixing-based methods are widely used. However, large-scale changes between categories cannot be detected accurately, as they cannot estimate the variance within individual categories, when the variances in the inhomogeneous areas are large. To solve these problems, a fusion method based on the frequency domain is proposed. It determines the reflectance of the high-resolution (HR) image using the frequency relationship between the HR image and the coarse-resolution one. It is faster and more accurate than the conventional spatial-domain-based fusion methods, as its operations are realized in the frequency domain. Both the large-scale textures and the small-scale textures can be preserved, even in the case of discern sudden or large-scale changes. Finally, experiments over simulated images and real satellite ones, using Landsat thematic mapper image and Sentinel-2 image, are carried out to demonstrate the performance of the proposed approach.