| HAYOLO: A novel object detection model for detecting defects in PCB |
| Paper ID : 1032-ICEEM2025 (R1) |
| Authors |
|
Ali Mohammed Elhenidy *1, Raid gaib2, Saif M. Alsabti2, Lara R. Al-Najjar2 11Computer Engineering and Control Systems Department. Faculty of Engineering, Mansoura University, Egypt.
2 Faculty of Engineering mansoura national university, Gamasa, Egypt. 2Technical College of Engineering, Al-Bayan University, Baghdad, Iraq |
| Abstract |
| Abstract— Automated printed circuit board (PCB) defect identification has grown in importance in the current electronics manufacturing industry, where speed and accuracy are crucial for maintaining product quality and cutting expenses. Real-time defect localisation is made possible by deep learning, especially object identification algorithms, which also reduce human error. In order to overcome the shortcomings in generalisability, robustness, and adaptability to unknown fault kinds, this work presents HAYOLO, an improved version of YOLOv12. To enhance multi-scale feature representation and detection accuracy, the suggested model integrates a Hierarchical Attention Fusion Module (HAFM), which combines spatial, channel, and coordinate attention methods. HAYOLO, which was trained using a three-fold cross-validation approach, achieves promising results with a mean Average Precision (mAP) of 98.4% while keeping a lightweight structure with just 2.2 million parameters. These findings demonstrate HAYOLO's potential for high-precision, real-time PCB fault identification in industrial settings, providing a workable and effective solution for smart production settings. |
| Keywords |
| YOLOv12; PCB ; defect detection |
| Status: Accepted |