Automatic Suit Lapel Types Detection and Classification Using Swin Transformer-Improved YOLOv5n

In the modern fashion industry, quickly and accurately classifying suit lapel types is essential for enhancing apparel design, production efficiency, and user experience on e-commerce platforms. This study proposes an improved YOLOv5n model with Swin-Transformer to address YOLOv5n's limitations in detail recognition and fine-grained features. By incorporating Swin-Transformer's global attention mechanism, the model significantly improves the detection of complex lapel types. A new dataset, SUIT_LAPEL_DATASET, was built, featuring four lapel styles: Notched, Peak, Shawl, and Collarless. Experiments show that the enhanced model reaches a [email protected] of 0.930, precision of 0.913, recall of 0.99, and F1 score of 0.90, outperforming the original YOLOv5n in complex backgrounds and varied lapel types. This approach advances the automatic detection of suit lapels. It supports clothing customization, virtual fitting systems, and intelligent e-commerce recommendations while demonstrating the advantages of combining convolutional networks with Transformer architectures.