Scholarlink

<p>Ensuring the reliability and high quality of electronic products necessitates accurate detection and classification of printed circuit board (PCB) defects. This study presents an enhanced defect detection model, TRSBi-YOLO. First, the incorporation of C3TR modules within the modified backbone structure improves multiscale feature extraction while significantly reducing computational complexity. Second, the integration of a simple parameter-free attention mechanism (SimAM) enhances feature representation without adding computational burden. Additionally, a bidirectional feature pyramid network (BiFPN) is employed in the neck section to optimize multi-level feature fusion and strengthen object detection capabilities. Experimental results demonstrate that the TRSBi-YOLO model achieves a remarkable mean average precision (mAP) of 98.1% with only 4.4G floating point operations (FLOPs) and 1.79M parameters. Compared to baseline model, TRSBi-YOLO shows substantial improvements, including a 2.40% increase in mAP, as well as enhancements in recall (0.62%), precision (0.73%), F1-score (1.12%), and accuracy (4.55%), along with a significant 11.99% increase in detection speed. These results demonstrate the effectiveness of the proposed TRSBi-YOLO model in improving both the detection and classification of PCB defects. The model offers a reliable and efficient solution for automated optical inspection (AOI), making it suitable for real-time quality control in industrial PCB manufacturing. Designed with deployment in mind, TRSBi-YOLO combines a lightweight architecture with high-speed performance, achieving an inference speed of 135.27 FPS. This makes it ideal for integration into high-throughput environments, including GPU-based systems and edge devices, supporting real-time and large-scale inspection tasks.</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

TRSBi-YOLO: Transformer based lightweight and high-performance model for PCB defects detection

  • 期刊论文

  • 发表日期：

    2025年8月27日

查看全文
  • Vinod Kumar Ancha,
  • Venkateswarlu Gonuguntla,
  • Ramesh Vaddi

摘要

Ensuring the reliability and high quality of electronic products necessitates accurate detection and classification of printed circuit board (PCB) defects. This study presents an enhanced defect detection model, TRSBi-YOLO. First, the incorporation of C3TR modules within the modified backbone structure improves multiscale feature extraction while significantly reducing computational complexity. Second, the integration of a simple parameter-free attention mechanism (SimAM) enhances feature representation without adding computational burden. Additionally, a bidirectional feature pyramid network (BiFPN) is employed in the neck section to optimize multi-level feature fusion and strengthen object detection capabilities. Experimental results demonstrate that the TRSBi-YOLO model achieves a remarkable mean average precision (mAP) of 98.1% with only 4.4G floating point operations (FLOPs) and 1.79M parameters. Compared to baseline model, TRSBi-YOLO shows substantial improvements, including a 2.40% increase in mAP, as well as enhancements in recall (0.62%), precision (0.73%), F1-score (1.12%), and accuracy (4.55%), along with a significant 11.99% increase in detection speed. These results demonstrate the effectiveness of the proposed TRSBi-YOLO model in improving both the detection and classification of PCB defects. The model offers a reliable and efficient solution for automated optical inspection (AOI), making it suitable for real-time quality control in industrial PCB manufacturing. Designed with deployment in mind, TRSBi-YOLO combines a lightweight architecture with high-speed performance, achieving an inference speed of 135.27 FPS. This makes it ideal for integration into high-throughput environments, including GPU-based systems and edge devices, supporting real-time and large-scale inspection tasks.