Noise in satellite images poses significant challenges, especially when they are riddled with salt-and-pepper noise this makes data analysis and interpretation difficult. However, when the noise density is high, traditional image processing methods find it difficult to recover important image features. The U-Net architecture is enhanced in this paper in a way that enables denoising with respect to visual clarity and details of the resulting image. Using batch normalization, dropout layers, and a custom loss function combining Mean Squared Error (MSE) and Structural Similarity Index (SSIM), the proposed approach improves the denoising process. Blurring effects are countered and image sharpness is enhanced using post-denoising application of unsharp masking. In his work, the model is trained with pairs of noisy and clean satellite images, and then evaluated on PSNR and SSIM metrics. Results show that such an approach significantly enhances image clarity and provides much needed contributions to improving the quality of satellite image data for use in practical applications such as environmental monitoring and urban planning.