Optimization of automatic classification for women’s pants based on the swin transformer model

Abstract In the post-pandemic era, integrating e-commerce and deep learning technologies is critical for the fashion industry. Automatic classification of women’s pants presents challenges due to diverse styles and complex backgrounds. This study introduces an optimized Swin Transformer model enhanced by the Global Attention Mechanism (GAM) to improve classification accuracy and robustness. A novel dataset, FEMPANTS, was constructed, containing images of five main trouser styles. Data preprocessing and augmentation were applied to enhance the model's generalization. Experimental results demonstrate that the improved model achieves a classification accuracy of 99.12% and reduces classification loss by 34.6%. GAM enhances the model's ability to capture global and local features, ensuring superior performance in complex scenarios. The research results not only promote the automation process in the fashion industry but also provide references for other complex image classification problems. This study highlights advancements in fashion e-commerce, offering practical applications for inventory management, trend analysis, and personalized recommendations, while paving the way for future innovations in deep learning-based image recognition.