A Survey of Deep Learning-based Pan-sharpening Techniques for Remote Sensing Images

Abstract

Remote sensing images are crucial for applications such as land-use mapping, environmental monitoring, and disaster management. Pan-sharpening enhances the spatial resolution of multispectral images by fusing them with high-resolution panchromatic images. Despite this, low spatial resolution can occur due to sensor limitations. To address this, image fusion methods, particularly pan-sharpening, have been developed to merge high-resolution and low-resolution images effectively. Recently, deep learning-based pan-sharpening techniques have gained prominence for achieving high-quality results. This survey offers a comprehensive overview of advancements in these techniques, reviewing and comparing various deep learning architectures, including autoencoder methods, generative adversarial networks (GANs), conditional GANs, convolutional neural networks (CNNs), and deep residual networks. We discuss the challenges, future directions, and advantages of deep learning in pan-sharpening while providing an in-depth analysis of state-of-the-art methods, their architectures, experimental results, evaluation metrics, and a comparative analysis of the surveyed techniques.