Multi-Frequency Fusion Attention Network for Pansharpening: Enhancing High-Resolution Multispectral Imaging in UAV Swarms

The problem of image fusion in unmanned aerial vehicles (UAVs) is of great significance in the field of remote sensing, but existing methods suffer from issues such as detail loss and frequency information confusion. With this in mind, this paper proposes a UAV image fusion scheme based on a Multi-Frequency Fusion Attention Network (MFFA), with the aim of generating high-resolution multispectral (HRMS) images by fusing low-resolution multispectral (LRMS) images with high-resolution panchromatic (PAN) images. The proposed method effectively addresses the issues of detail loss and frequency information confusion present in traditional methods through MFFA and the Spatial Detail Enhancement Module (SDEM). Specifically, the MFFA module utilizes the Discrete Wavelet Transform (DWT) to separate the frequency components of the PAN image and achieves precise fusion of different frequency information through a frequency attention triplet. The SDEM module further enhances the spatial detail information of the image. Additionally, a multiscale fusion framework is employed to progressively fuse information at different scales, improving the quality of the final generated image. Experimental results on multiple datasets show that the proposed method demonstrates excellent fusion performance and generalizability.