A Semantic–Visual Reasoning Framework for Guangdong Temple Fairs

Digital technologies have significantly advanced the visual archiving of intangible cultural heritage (ICH).However, existing methods often struggle to interpret the complex social structures embedded within these activities.This limitation is particularly evident in the digitization of cultural heritage in Guangdong.While state-of-the-art visual language models (VLMs) can identify superficial visual elements-such as clothing, crowds, and architectural forms-they fail to recognize the underlying ritual rules, clan organizations, and spatial hierarchies.To bridge this cognitive gap, this paper proposes a Semantic-Visual Reasoning Framework (SVRF).This framework combines domain-specific knowledge graphs (KGs) with VLM-based visual extraction to ensure semantic consistency in scene understanding.Focusing on the intangible cultural heritage of Guangdong temple fairs, we constructed an ontology that explicitly models key sociological elements, including participants, sacred spaces, and ritual sequences.This ontology serves as a semantic constraint layer, guiding visual analysis.Through a qualitative case study of the Foshan Ancestral Temple Fair, we illustrate how SVRF reduces misinterpretations in high-context cultural scenes by integrating visual evidence with social logic.This research provides a novel computational approach for interpreting the multi-layered social dynamics of regional heritage activities, facilitating a shift beyond static archiving towards semantic interpretation.