A Generative Adversarial Network to Upscale the Resolution of Low-Resolution Galaxy Images

High-resolution galaxy images play a crucial role in astronomy and astrophysics, enabling detailed morphological analysis, the identification of faint structures, and distance measurements. However, limitations in telescope technologies, atmospheric conditions, and the vast distances of observed objects often result in low-resolution images. In this research, we leveraged advanced deep-learning methods to enhance the resolution of low-resolution galaxy images. We introduced a Generative Adversarial Network (GAN) specifically developed to upscale galaxy images taken at suboptimal resolutions, with the Enhanced Super Resolution Generative Adversarial Network (ESRGAN) serving as the foundational model. Our self-attention multi-scale GAN architecture utilized deep learning to produce super-resolution images by learning from a dataset of high-resolution galaxy images. Additionally, we incorporated an Explainable AI technique to open the black box of our GAN model, providing insights into its intrinsic decision processes and feature learning mechanisms. This advanced approach not only enhanced image resolution but also improved model transparency and built trust in the model's predictions. Our results demonstrated a significant improvement in the Peak Signal-to-Noise Ratio and Structural Similarity Index Measure and provided a better understanding of the model's operations, which is highly important for scientific analysis and validation in astronomical research.