Deep Learning for Automated Recognition and Digital Reconstruction of Ceramic Motifs

Ceramic motifs embody the core aesthetic, technological, and cultural values of numerous civilizations. However, environmental degradation, physical damage, and limited conservation resources have resulted in the loss of many artifacts. Therefore, the digital preservation and reconstruction of ceramic patterns is crucial for protecting cultural heritage and supporting research. This study proposes an integrated computational framework that combines deep learning-based recognition with generative reconstruction techniques. Convolutional neural networks (CNNs) and visual transformers (ViTs) were used for automatic pattern recognition. In contrast, generative adversarial networks (GANs) and latent diffusion models (LDMs) were used to reconstruct damaged or missing patterns. The dataset was curated from museum collections, archaeological archives, and field surveys to ensure both diversity and authenticity. This study used a curated dataset drawn from museum collections, archaeological archives, and field surveys. Results showed that transformer-based recognition achieved over 93% accuracy. At the same time, diffusion-based reconstructions produced clearer and more stylistically coherent patterns than GAN-generated patterns, as confirmed by objective metrics and expert evaluation. These findings highlight the potential of artificial intelligence to supplement conservation practices and create high-quality digital replacements for ceramic heritage, offering scalable tools for preservation and providing new resources for comparative studies in archaeology and art history.