Adaptive High-Resolution PCB Defect Detection Using YOLO11n with Tile-Based Processing and Global Coordinate Reconstruction

This paper presented a software-based framework for high-resolution PCB defect inspection that combines YOLO11n with adaptive overlapping tiling, coordinate reconstruction and global class-aware NMS. The method addresses the loss of small-defect information that occurs when a high-resolution board is directly reduced to the detector input size. The baseline model achieved 96.2% precision, 93.8% recall, 93.5% mAP@50 and 69.9% mAP@50–95, and the fine-tuned model achieved 98.6% precision, 85.6% recall, 84.6% mAP@50 and 48.1% mAP@50–95 on a 75-image test set. High precision does not imply complete coverage: the lower recall, concentrated in mouse-bite and spurious-copper defects, and the gap between mAP@50 and mAP@50–95 show that missed detections and box localization are the main remaining problems.

[1]   W. Huang and P. Wei, “A PCB dataset for defects detection and classification,” arXiv preprint arXiv:1901.08204, 2019. [2]   J. Redmon, S. Divvala, R. Girshick, and A. Farhadi, “You only look once: Unified, real-time object detection,” in Proc. IEEE Conf. Computer Vision and Pattern Recognition (CVPR), 2016, pp. 779–788. [3]   G. Jocher and J. Qiu, “Ultralytics YOLO11,” 2024. [Online]. Available: https://docs.ultralytics.com/models/yolo11 [4]   F. O. Ünel, B. O. Özkalaycı, and C. Çiğla, “The power of tiling for small object detection,” in Proc. IEEE/CVF CVPR Workshops, 2019. [5]   F. C. Akyon, S. O. Altinuc, and A. Temizel, “Slicing aided hyper inference and fine-tuning for small object detection,” in Proc. IEEE Int. Conf. Image Processing (ICIP), 2022, pp. 966–970. [6]   A. Neubeck and L. Van Gool, “Efficient non-maximum suppression,” in Proc. Int. Conf. Pattern Recognition (ICPR), 2006, pp. 850–855.