Restoring damaged archaeological artifacts is important for the preservation of cultural heritage and for the possibility of historical research. This paper introduces a new approach that combines two deep learning models, U-Net and CycleGAN, to restore images of degraded artifacts. U-Net is used for the first restoration, which reduces noise and local damage such as cracks and occlusion, while CycleGAN refines these outputs by maintaining stylistic consistency and enhancing intricate details. The framework is trained on the dataset provided by Zhao (Mendeley Data V1, 2023 [1]), which includes photographic images of painted pottery artifacts. To improve robustness, artificial degradation techniques such as scratches, occlusions, and cracks were applied. The methodology involves simulating realistic damage patterns over undamaged artifact images and sequentially training the models before their evaluation using objective metrics like Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM) in addition to the expert visual inspection. The combined U-Net + CycleGAN approach yielded superior performance over standalone models with a PSNR of 30.7 dB and an SSIM of 0.92, showing high fidelity and perceptual similarity. However, challenges remain in handling highly intricate artifact details and variations in real-world degradation patterns. Addressing these issues will be crucial for practical deployment in cultural heritage preservation, museum digitization, and automated artifact restoration.