基于HRA-YOLO的蓝印花布纹样分类和识别OA

To address the challenges of limited technical diversity,low accuracy and weak generalization in digital classification and recognition of the intangible cultural heritage blue calico,an improved pattern classification and recognition algorithm of blue calico named HRA-YOLO based on YOLOv11n was proposed.Modified HS-FPN based on Feature Pyramid Network(FPN)was constructed for the model,accuracy and robustness of the classification and recognition were improved,the lightweight was realized.Adaptive Spatial Feature Fusion(ASFF)was introduced to improve detection head,the capability of processing different scale blue calico pattern was improved.Receptive Field Attention Convolution(RFAConv)was incorporated to boost the model performance without increasing computational costs and parameters.The conventional loss function was replaced by an angle-aware and geometry-constrained SIoU loss function to accelerate convergence and improve detection efficacy.Additionally,blue calico pattern dataset was constructed and expanded through diverse data augmentation strategy.The test results demonstrated that compared to YOLOv11n,mAP@0.5 and mAP@0.5:0.95 of HRA-YOLO were increased 4.1 percentage points and 2.8 percentage points respectively,while the parameters was reduced 18.2％.Compared to seven YOLO series algorithms,including YOLOv3-T,YOLOv5n,YOLOv6n,YOLOv8n,YOLOv8ghost,YOLOv9-T and YOLOv10n,the classification and recognition accuracy of HRA-YOLO algorithm was superior to other algorithms,and the overall detection performance was the best.It is considered that HRA-YOLO algorithm is suitable for digital classification and recognition tasks of blue calico patterns.