Scholarlink

<p>Accurate and rapid detection of surgical instruments is crucial for robot-assisted laparoscopic surgery. Although machine vision-based methods have advanced significantly in instrument detection, achieving high accuracy remains challenging. To address this, we introduce Tool-YOLO, an enhanced variant of YOLOv10, incorporating three innovative modules: the Occlusion Weighted Fusion Module (OWFM), the C2f-Cross-stage Partial connections with Convolutional Single-head attention Hybrid module (C2f-CSH), and the CSPPyramid Net Fusion (CSPPNF). The OWFM integrates occlusion-aware SEAM with channel attention to enhance multi-scale feature fusion and performance under occlusion conditions. The C2f-CSH module improves upon the conventional C2f module by replacing its residual blocks with a CSH module that employs Convolutional GLU (CGLU) from TransNeXt, supplanting the single-head attention FFN block in SHvit. The CSPPNF module synergizes multi-scale convolutions, feature separation and reorganization, and residual connections to bolster feature diversity and representation. On the M2CAI16-Tool-Locations dataset, Tool-YOLO achieves a detection accuracy of 95.9%, an mAP@0.5 of 97.4%, and an mAP@0.5:0.95 of 0.643, surpassing the baseline YOLOv10 by 2.2%, 3.3%, and 7.4%, respectively. On the Endoscapes-CVS201 dataset, Tool-YOLO records an mAP@50 of 96.3%, demonstrating robust performance compared to existing methods. Furthermore, the model achieves an mAP@0.5:0.95 of 42.4% on the COCO2017 dataset, indicating strong generalization capabilities. These results underscore Tool-YOLO’s potential as a high-performance solution for surgical instrument detection and broader object detection applications.</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

Tool-YOLO: a target detection network based on feature extraction and feature fusion

  • 期刊论文

  • 发表日期：

    2025年10月10日

查看全文
  • Guanqun Guo,
  • Li Zhang,
  • Boqiang Jia,
  • Jiayu Zhang,
  • Wenjie Wang

摘要

Accurate and rapid detection of surgical instruments is crucial for robot-assisted laparoscopic surgery. Although machine vision-based methods have advanced significantly in instrument detection, achieving high accuracy remains challenging. To address this, we introduce Tool-YOLO, an enhanced variant of YOLOv10, incorporating three innovative modules: the Occlusion Weighted Fusion Module (OWFM), the C2f-Cross-stage Partial connections with Convolutional Single-head attention Hybrid module (C2f-CSH), and the CSPPyramid Net Fusion (CSPPNF). The OWFM integrates occlusion-aware SEAM with channel attention to enhance multi-scale feature fusion and performance under occlusion conditions. The C2f-CSH module improves upon the conventional C2f module by replacing its residual blocks with a CSH module that employs Convolutional GLU (CGLU) from TransNeXt, supplanting the single-head attention FFN block in SHvit. The CSPPNF module synergizes multi-scale convolutions, feature separation and reorganization, and residual connections to bolster feature diversity and representation. On the M2CAI16-Tool-Locations dataset, Tool-YOLO achieves a detection accuracy of 95.9%, an mAP@0.5 of 97.4%, and an mAP@0.5:0.95 of 0.643, surpassing the baseline YOLOv10 by 2.2%, 3.3%, and 7.4%, respectively. On the Endoscapes-CVS201 dataset, Tool-YOLO records an mAP@50 of 96.3%, demonstrating robust performance compared to existing methods. Furthermore, the model achieves an mAP@0.5:0.95 of 42.4% on the COCO2017 dataset, indicating strong generalization capabilities. These results underscore Tool-YOLO’s potential as a high-performance solution for surgical instrument detection and broader object detection applications.