Intelligent Recognition of Ethnic Costumes Using YOLOv11: A Deep Learning Framework for Cultural Heritage Preservation

The rapid digitization of cultural heritage creates new opportunities to preserve the visual and symbolic richness of ethnic traditions. However, accurate recognition of ethnic costumes remains challenging due to complex textures, overlapping patterns, and high inter-group similarity. This study proposes an intelligent recognition framework based on the YOLOv11 architecture for the digital preservation of multi-ethnic attire in Lijiang, China. By integrating a C2PSA spatial attention mechanism with multi-scale feature fusion, the model enhances discrimination of fine-grained textile structures under complex visual conditions. A large-scale dataset containing 4,974 images from 55 ethnic branches was constructed for evaluation. Experimental results demonstrate that the proposed method achieves an mAP@0.5 of 98.416% and a recall of 95.923%, significantly outperforming YOLOv5 and YOLOv4 (p < 0.001). With only 5.8 million parameters and 16.2 GFLOPs, the model enables efficient real-time deployment, contributing a robust AI-driven solution for cultural heritage informatics.