Scholarlink

<p>With the rapid development of the electronics industry, PCB defect detection is a core aspect of quality control in electronic manufacturing. Existing detection models often focus on optimizing the precision for small target detection but neglect the improvement of detection speed, leading to a core contradiction where precision and efficiency cannot be balanced. To address this issue, we propose a lightweight and high-precision PCB defect detection model, YOLO-VCL, based on YOLOv11. First, the C3K2 module is creatively combined with the vision state space duality (VSSD) block in the backbone network to cooperatively optimize precise local feature capture and global context modeling, while enhancing detection accuracy and controlling computational cost. Second, the neck network employs a context anchor attention (CAA) module to enhance the high-level screening feature pyramid network (HSFPN). Through long-distance context capturing and selective feature fusion, it strengthens the correlation of multi-scale defect features, thus improving the model’s detection precision and computational efficiency. Finally, a lightweight shared convolution detection head (LSCD-head) is proposed, combining group normalization and scale-adaptive adjustment to reduce parameters and computational cost while ensuring detection accuracy. Experimental results on the PCB defect dataset show that the mAP@0.5 and mAP@0.5:0.95 of the YOLO-VCL model reach 98.4% and 70.3%, respectively, improving by 0.8% and 2.5% over the baseline model; at the same time, the model’s parameters and computation decrease to 1.86 M and 5.7 GFLOPs, representing a reduction of 28.1% and 9.5% compared to the baseline model, while maintaining an inference speed of 149.77 FPS, meeting the real-time detection needs of industrial production lines.</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

YOLO-VCL: a lightweight and high-precision PCB defect detection network model

  • 期刊论文

  • 发表日期：

    2026年2月10日

查看全文
  • Wenbo Zhang,
  • Wei Chen,
  • Yunfeng Ni,
  • Shouhua Wang,
  • Bei Dang

摘要

With the rapid development of the electronics industry, PCB defect detection is a core aspect of quality control in electronic manufacturing. Existing detection models often focus on optimizing the precision for small target detection but neglect the improvement of detection speed, leading to a core contradiction where precision and efficiency cannot be balanced. To address this issue, we propose a lightweight and high-precision PCB defect detection model, YOLO-VCL, based on YOLOv11. First, the C3K2 module is creatively combined with the vision state space duality (VSSD) block in the backbone network to cooperatively optimize precise local feature capture and global context modeling, while enhancing detection accuracy and controlling computational cost. Second, the neck network employs a context anchor attention (CAA) module to enhance the high-level screening feature pyramid network (HSFPN). Through long-distance context capturing and selective feature fusion, it strengthens the correlation of multi-scale defect features, thus improving the model’s detection precision and computational efficiency. Finally, a lightweight shared convolution detection head (LSCD-head) is proposed, combining group normalization and scale-adaptive adjustment to reduce parameters and computational cost while ensuring detection accuracy. Experimental results on the PCB defect dataset show that the mAP@0.5 and mAP@0.5:0.95 of the YOLO-VCL model reach 98.4% and 70.3%, respectively, improving by 0.8% and 2.5% over the baseline model; at the same time, the model’s parameters and computation decrease to 1.86 M and 5.7 GFLOPs, representing a reduction of 28.1% and 9.5% compared to the baseline model, while maintaining an inference speed of 149.77 FPS, meeting the real-time detection needs of industrial production lines.