Scholarlink

<p>Stitch code generation technology plays a crucial role in enhancing designers’ efficiency during the production process and shortening the production cycle. However, existing methods face the problem of low accuracy in generating low-frequency stitch codes, which are vital for complex knitting patterns. In this paper, we propose LFSCG-E-Net, an enhanced model based on InverseKnit, aiming to improve the generation accuracy of low-frequency stitch codes. By integrating a feature pyramid network within RefinerNet, we improve the texture detail enhancement of input images. Furthermore, we introduce residual attention block with convolutional block attention modules into InferNet to boost feature extraction capabilities. Additionally, a pyramidal feature hierarchy module is developed to extract and merge multi-scale feature maps by combining the spatial pyramid depth and atrous spatial pyramid pooling. To tackle class imbalance, we incorporate the Dice coefficient into the loss function. Experiments on a public dataset demonstrate that our model achieves an overall accuracy of 94.9% and a foreground accuracy of 82.87%, outperforming state-of-the-art methods, especially in low-frequency stitch code generation. This work not only enhances the precision of stitch code generation but also contributes to the automation of knitting pattern design, meeting the industry’s demand for efficiency and cost-effectiveness. The source code is available at <a href="https://github.com/1033216625/LFSCG-E-Net-">https://github.com/1033216625/LFSCG-E-Net-</a>.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhancing low-frequency stitch code generation for knitted fabrics: an LFSCG-E-Net approach

  • 期刊论文

  • 发表日期：

    2025年3月8日

查看全文
  • Jinxing Liang,
  • Kaifang Han,
  • Dongsheng Li,
  • Ruixin Gao,
  • Jiajia Peng,
  • Tao Peng,
  • Xinrong Hu

摘要

Stitch code generation technology plays a crucial role in enhancing designers’ efficiency during the production process and shortening the production cycle. However, existing methods face the problem of low accuracy in generating low-frequency stitch codes, which are vital for complex knitting patterns. In this paper, we propose LFSCG-E-Net, an enhanced model based on InverseKnit, aiming to improve the generation accuracy of low-frequency stitch codes. By integrating a feature pyramid network within RefinerNet, we improve the texture detail enhancement of input images. Furthermore, we introduce residual attention block with convolutional block attention modules into InferNet to boost feature extraction capabilities. Additionally, a pyramidal feature hierarchy module is developed to extract and merge multi-scale feature maps by combining the spatial pyramid depth and atrous spatial pyramid pooling. To tackle class imbalance, we incorporate the Dice coefficient into the loss function. Experiments on a public dataset demonstrate that our model achieves an overall accuracy of 94.9% and a foreground accuracy of 82.87%, outperforming state-of-the-art methods, especially in low-frequency stitch code generation. This work not only enhances the precision of stitch code generation but also contributes to the automation of knitting pattern design, meeting the industry’s demand for efficiency and cost-effectiveness. The source code is available at https://github.com/1033216625/LFSCG-E-Net-.