Scholarlink

<p>Industrial cutting tools are widely used in modern manufacturing, and deep learning-based methods have become common for their classification. However, existing approaches often face issues like low accuracy and poor generalization. To address this, an improved detection algorithm based on YOLOv11 is proposed. The Cross Stage Partial Transformer Block is redesigned with Transformer structures to enhance global feature extraction, and the original C3k2 module is replaced by an improved Multi-Scale Convolution Block for better multi-scale fusion. The Spatial-Channel Squeeze Excitation Module is further introduced to enhance attention across scales, while the existing SIoU loss function is employed as a complementary technique to improve localization accuracy. The results showed that compared with the baseline, the optimized model size decreased to 2.37M (8.1% decrease), mAP@0.5 reached 94.98% (1.9% increase), and accuracy increased to 95.71% (5.8% increase).</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

TMSS-YOLO: An industrial tool detection method based on improved YOLOv11

  • 期刊论文

  • 发表日期：

    2025年9月23日

查看全文
  • Guan Yang,
  • ZhiXin Zhang,
  • HuaDong Sang,
  • Hao Tang,
  • Miao Wang

摘要

Industrial cutting tools are widely used in modern manufacturing, and deep learning-based methods have become common for their classification. However, existing approaches often face issues like low accuracy and poor generalization. To address this, an improved detection algorithm based on YOLOv11 is proposed. The Cross Stage Partial Transformer Block is redesigned with Transformer structures to enhance global feature extraction, and the original C3k2 module is replaced by an improved Multi-Scale Convolution Block for better multi-scale fusion. The Spatial-Channel Squeeze Excitation Module is further introduced to enhance attention across scales, while the existing SIoU loss function is employed as a complementary technique to improve localization accuracy. The results showed that compared with the baseline, the optimized model size decreased to 2.37M (8.1% decrease), mAP@0.5 reached 94.98% (1.9% increase), and accuracy increased to 95.71% (5.8% increase).