Lightweight visual feature recognition of Tibetan Opera costumes: framework design and implementation

To promote the digital preservation and dissemination of Tibetan Opera, a unique form of intangible cultural heritage, this study addresses key challenges such as limited sample availability, intricate costume patterns, variable object scales, and frequent occlusions. Building upon the YOLOv9t architecture, we propose an improved lightweight recognition model, termed MBC-YOLO. This model integrates the Mamba module into the neck component to enhance the modeling of long-range dependencies, and designs a Bidirectional Coupled Feature Pyramid Network (BC-FPN) to improve cross-scale feature fusion efficiency. Furthermore, a lightweight strategy is adopted to support deployment on edge devices. Experimental results on a self-constructed, small-sample Tibetan Opera costume image dataset demonstrate that the proposed model achieves a mAP50-95 of 65.6%, exhibiting enhanced robustness in scenarios with complex backgrounds and multiple overlapping targets. This work provides an efficient and practical solution for intelligent recognition of traditional costume heritage under small-sample conditions.