Abstract:

Image segmentation methods have been used in the field of recognition and segmentation of traditional embroidery patterns. However, due to the characteristics of complex edges, cumbersome details and diverse types of embroidery patterns, traditional image segmentation methods are difficult to meet the requirements of high precision and high efficiency in practical applications. So, this paper presents a dual-stage embroidery pattern recognition and segmentation method based on a YOLO and U-Net cascade. In the first stage, the YOLO algorithm is employed for object detection, quickly and accurately locating the embroidery patterns within the image. In the second stage, an enhanced U-Net algorithm is used for semantic segmentation. The U-Net encoder structure is improved by incorporating a ResBlock-CBAM module as the backbone, enhancing the effectiveness of feature extraction, and integrating an ASPP module for feature enhancement to ensure effective extraction and fusion of various features. This dual-stage cascade network captures the fine details and contextual information of embroidery images, enabling precise segmentation that preserves complex edges and details. The experimental results show that the algorithm in this study reaches 0.8584 and 0.8376 in the evaluation indicators such as Dice and MioU, respectively, and the accuracy rate reaches 84.53%, which is significantly better than other advanced segmentation algorithms. At the same time, this paper establishes an "embroidery intelligent recognition and segmentation" system to achieve efficient and automatic extraction and processing of embroidery patterns. This method not only provides technical support for the digital preservation and transmission of embroidery patterns, but also paves the way for the modernization and customization of embroidery design.

Key words: