Artificial Intelligence–Driven Extraction and Innovative Design of Cultural Factors in Miao Costume Patterns from Qiandongnan

In the context of global efforts to preserve cultural diversity and the expanding role of artificial intelligence in creative industries, this study presents an artificial intelligence–driven framework for extracting and innovatively reinterpreting cultural factors embedded in the costume patterns of the Miao people in Qiandongnan, China. By integrating ethnographic fieldwork with computational analysis, the research systematically examines the visual, semantic, and symbolic dimensions of Miao pattern culture. A comprehensive database (MiaoPattern-DB) of 1,600 high-resolution images of embroidery, weaving, and silver ornaments was compiled from 16 Miao-inhabited counties, including Kaili, Leishan, Taijiang, Liping, and Congjiang. We applied multimodal fusion techniques—feature detection with YOLOv9, hierarchical representation using Swin Transformer, and cross-modal semantic alignment via CLIP—to identify five principal cultural factors (pattern, structure, color, semantics, and morphology) and seventeen sub-factors. The database maps relationships among visual form, cultural meaning, and regional identity, and these relationships were further formalized in a Neo4j-based knowledge graph. Generative design tools (ComfyUI and Stable Diffusion) were employed to reconstruct traditional motifs and translate them into contemporary design products, enabling creative revitalization of ethnic visual heritage. This study bridges computational design and intangible cultural heritage preservation, offering a scalable methodological paradigm for technology-enabled research in ethnic arts and contributing to the sustainable transmission and contemporary renewal of Miao visual culture.