Intelligent detection of Shu embroidery fabric stitches based on DDW YOLOv8

Abstract

The intelligent recognition of Shu embroidery fabric stitches is particularly crucial for the advancement and inheritance of Shu embroidery. This paper completes the pioneering effort to amalgamate deep learning methodologies with Shu embroidery fabric stitches, presenting a novel recognition approach based on Dynamic Snake Asymptotic Feature Pyramid Network–Dual Attention Network–WIoU YOLOv8 (DDW YOLOv8) for Shu embroidery fabric stitches. In the feature extraction stage, the Simple, Parameter-Free Attention Module (SimAM) and the Selective Kernel Networks Attention (SKAttention) are integrated into the C2f module and arranged in sequence, enhancing the model’s capacity to extract critical features across multiple scales. To prevent inadequate fusing of stitch method features during the feature fusion phase, the Dynamic Snake Asymptotic Feature Pyramid Network (DSAFPN) is introduced, enabling the model to accommodate stitch features and concurrently realize multilevel feature fusion. In addition, the improvement of the WIoU can increase the precision of the model when recognizing closely arranged stitches. Finally, DDW YOLOv8 is proposed. The experimental findings demonstrate that DDW YOLOv8 has achieved significant improvements over both YOLOv8 and YOLOv11, with increases of 4.63% and 3.2% in mAP@0.5, and 4.02% and 7.18% in mAP@0.5:0.95, respectively. Notably, DDW YOLOv8 has successfully identified the stitches of Shu embroidery fabric for the first time, providing robust support for the inheritance and advancement of Shu embroidery while simultaneously laying the groundwork for the intelligent recognition of Shu embroidery fabric stitches.