A study on Guangxi Zhuang traditional costume patterns based on GhostNetV2 and attention mechanism

Economic development has led to the commercialization of ethnic attire, prioritizing popular aesthetics over cultural preservation, which complicates the recognition of traditional Zhuang patterns. Deep learning methods, while useful, often require large datasets and high computational resources, limiting their use in remote areas. This study optimizes the YOLOv5 model by integrating the GhostNetV2 algorithm with SimAM and SEAttention mechanisms, improving efficiency for devices with limited computational power. The optimized model reduces computational cost by one-third, cuts 2,152,806 parameters, and enhances feature detail capture through attention mechanisms.