Machine Learning and Deep Learning for Cultural Heritage Conservation: A Bibliometric and Task-Oriented Review

With the rapid advancement of Artificial Intelligence (AI) technologies, Machine Learning (ML) and Deep Learning (DL) have become pivotal methods for driving the digital documentation, restoration, preservation, and preventive conservation of Cultural Heritage (CH). This paper constructs an integrated “data + technology + task” framework tailored for CH scenarios. It employs a combination of bibliometric analysis and systematic content study based on relevant literature published between 2011 and 2025. First, publication trends, sources of publication, global collaboration networks, and topic modeling reveal the overall landscape and evolutionary path of research on the digitization and intelligent transformation of CH. Subsequently, beginning with ML and DL systems, it summarizes classic workflows and outlines their applications in CH conservation. Concurrently, integrating topic modeling, existing research is categorized into three themes based on task attributes: Recognition, Reconstruction and Virtual Restoration, and Monitoring and Prediction. Representative literature, typical tasks, and technological trends within each theme are systematically outlined. Distinct from existing reviews, this study introduces a unified data technology task framework that explicitly links AI model paradigms to heritage-specific constraints. Moving forward, by constructing high-quality heritage datasets, enhancing model interpretability, and exploring cross-model fusion approaches, AI technologies hold promise to play a more reliable and sustainable role in CH conservation, risk management, and digital dissemination.