8th ISF

COMP033 - AstroDS U-NET/GAN: Revisiting U-NET for Astronomical Image Denoising Based on SSIM loss and GAN

Advancements in astronomical technologies have led to significant innovations that can enhance the resolution of deep-sky objects within a short period. For example, the James Webb Space Telescope can generate over 40 TB of high-resolution images daily. However, these images often consist of noises, which lead to colour inconsistencies. Despite the various denoising methods used nowadays, they can still be improved. Particularly, due to the large number of data produced daily, which exceeds the capacity for manual analysis. To overcome this challenge, machine learning are employed to reduce noises in astronomical images. Generative Adversarial Networks (GAN) and Structural Similarity Index (SSIM) models are used to improve the image quality. These combination of machine learning models can achieve a revolutionary efficiency by requiring less training time and data, and enhancing the effectiveness of the outcomes.

Show More