Revolutionizing Pan Sharpening in Remote Sensing with Cutting-Edge Deep Learning Optimization

Deep Learning (DL) techniques have significantly advanced the area of remote sensing, particularly in the enhancement of pan sharpening-a process that combines high- resolution (HR) panchromatic images with lower-resolution (LR) multispectral images to produce high-quality, detailed imagery. Traditional pan sharpening methods, which relied on component substitution and multiresolution analysis, often struggled with balancing spatial and spectral quality, leading to artifacts and degraded image quality. DL has further enhanced remote sensing applications, including crop classification, image change detection, and environmental monitoring. Notable contributions in this domain include use of deep neural networks for citrus crop classification and employment of deep forest modules for change detection. Additionally, integrating multi- source data has shown promising results, explored the benefits of combining various data sources for improved classification and monitoring. Recent advancements highlight the application of DL models across diverse remote sensing tasks: SFRNet-MLFF for hyperspectral image sub-pixel mapping, Trans-MAD for change detection, MCAT-UNet for image segmentation, PV-Unet for photovoltaic panel extraction, LGDA for object detection, and M3SPADA for land cover mapping. These models illustrate the wide-ranging capabilities of DL in enhancing accuracy and efficiency in remote sensing applications. Specifically, the PV-Unet model, featuring an encoder-decoder architecture, demonstrated exceptional performance in photovoltaic panel extraction with an accuracy of 95.77%, precision of 99.09%, recall of 94.20%, F1 score of 96.59%, and Intersection over Union (IoU) of 93.40%.