The AI-driven transformation of cultural heritage image recognition: a scientometric analysis of knowledge transfer, evolving fronts, and ethical challenges

This study performs a scientometric analysis of 479 publications from the Web of Science to map the evolution of deep learning in cultural heritage image recognition. Results reveal a fundamental shift in knowledge transfer: the ‘cross-disciplinary citation lag’ has compressed from a historical eight years to just 1–3 years, signifying a qualitative leap in the field’s absorptive capacity. This acceleration parallels an exponential growth in research output since 2020. Structurally, the research agenda has deepened from perception-based ‘image recognition’ to cognitive frontiers like ‘knowledge graphs’ and ‘generative adversarial networks’. Crucially, this study provides quantitative evidence that ‘ethics’ has evolved from a peripheral issue into a central ‘motor theme’. This finding highlights that profound ethical challenges–specifically data sovereignty and digital colonialism–are emerging as core drivers of the field’s future trajectory, necessitating a balance between technological advancement and ethical responsibility.