Article: Optimisation of intelligent recognition teaching system for Miao costume patterns integrating YOLOv5 Journal: International Journal of Information and Communication Technology (IJICT) 2025 Vol.26 No.49 pp.57 - 74 Abstract: This study builds a high-quality Miao pattern dataset, then applies label smoothing and mosaic data augmentation. To maximise multi-scale feature fusion, the spatial pyramid pooling fast (SPPF) module is utilised. Increasing the precision of bounding box regression and small target recognition, the focal loss and complete IoU Loss algorithms are combined. A web-based visual teaching platform is created with features for displaying cultural knowledge and inferring models. The research results indicate that, the enhanced YOLOv5 model outperforms comparable models like faster R-convolutional neural network and YOLOv4 with mean average precision@0.5 of 89.6% and mAP@0.5:0.95 of 61.5% on the test set. Compared to the original YOLOv5s, it has increased by 5.4% and 7.2% respectively. Meantime, the recall rate improvement in small pattern detection is greater than 6%, which is better than that of baseline models such as YOLOv4. The data confirm deep learning's potential in high-precision ethnic culture recognition and instruction. Inderscience Publishers - linking academia, business and industry through research

Title: Optimisation of intelligent recognition teaching system for Miao costume patterns integrating YOLOv5

Authors: Qiong Luo; Xubing Xu; Jan Zhou

Addresses: College of Fashion and Art Design, Donghua University, Shanghai, 200000, China; School of Art and Design, Anhui Institute of Information Technology, Wuhu, 241000, China ' College of Fashion and Art Design, Donghua University, Shanghai, 200000, China ' School of Art and Design, Anhui Institute of Information Technology, Wuhu, 241000, China

Abstract: This study builds a high-quality Miao pattern dataset, then applies label smoothing and mosaic data augmentation. To maximise multi-scale feature fusion, the spatial pyramid pooling fast (SPPF) module is utilised. Increasing the precision of bounding box regression and small target recognition, the focal loss and complete IoU Loss algorithms are combined. A web-based visual teaching platform is created with features for displaying cultural knowledge and inferring models. The research results indicate that, the enhanced YOLOv5 model outperforms comparable models like faster R-convolutional neural network and YOLOv4 with mean average precision@0.5 of 89.6% and mAP@0.5:0.95 of 61.5% on the test set. Compared to the original YOLOv5s, it has increased by 5.4% and 7.2% respectively. Meantime, the recall rate improvement in small pattern detection is greater than 6%, which is better than that of baseline models such as YOLOv4. The data confirm deep learning's potential in high-precision ethnic culture recognition and instruction.

Keywords: Miao costume patterns; YOLOv5; target detection; transformer attention mechanism; intelligent teaching system.

DOI: 10.1504/IJICT.2025.151063

International Journal of Information and Communication Technology, 2025 Vol.26 No.49, pp.57 - 74

Received: 29 Sep 2025
Accepted: 31 Oct 2025
Published online: 12 Jan 2026 *

Title: Optimisation of intelligent recognition teaching system for Miao costume patterns integrating YOLOv5

Keep up-to-date