Space imagery is critical for astronomical
research and satellite observations, but it often suffers from
noise and artifacts, such as streaks caused by cosmic rays and
sensor imperfections. Traditional denoising techniques
frequently compromise essential features like stars, leading to a
loss of valuable information. This paper introduces Stellar
Cleanse, a pioneering framework that leverages CosmicNet for
superior denoising of space imagery. CosmicNet integrates
several advanced techniques: frequency domain preprocessing
with Fourier Transform and low-pass filtering, a customized
convolutional neural network (CNN) for residual noise learning,
attention mechanisms for streak detection and preservation of
stars, Total Variation (TV) Minimization for edge retention, and
a Generative Adversarial Network (GAN) for final image
refinement. Extensive experimental evaluation demonstrates
that CosmicNet significantly outperforms traditional methods
in both quantitative metrics (PSNR, SSIM) and qualitative
assessments, effectively reducing noise while preserving critical
celestial features. This groundbreaking approach not only
enhances image clarity but also holds promise for broader
applications in astronomical imaging and beyond