Defect detection method of printed circuit boards based on EDF-YOLOv10

To address the challenges of inadequate feature representation for small objects and slow model convergence in printed circuit board (PCB) defect detection, this paper proposes an improved YOLOv10 algorithm and develops a real-time detection system with a co-optimized hardware and software architecture. The efficient channel attention (ECA) mechanism is used to enhance the ability of the model to extract key channel features; the dynamic snake convolution (DSConv) in the backbone strengthens the model's capacity to recognize the geometric structures of small targets through deformable kernels and multi-directional feature fusion; the Focaler-CIoU loss emphasizes samples with low intersection over union (IoU) values to boost hard sample learning and improve convergence efficiency. To simulate real-world industrial environments, multiple data augmentation strategies are utilized to expand the PKU-Market-PCB dataset, thereby enhancing the model's generalization and robustness in complex scenarios. Experimental results demonstrate that the proposed EDF-YOLOv10 achieves mAP@0.50 of 90.6% and mAP@0.50:0.95 of 48.4% on the experimental dataset, representing improvements of 3.0 and 1.6 percentage points over the baseline, respectively. Furthermore, We also develope a real-time interactive detection system for identifying PCB defects. This system utilizes industrial cameras, a controllable light source, and a graphical user interface developed with the PyQt5 framework, employing the EDF-YOLOv10 model. Our approach serves as a methodological reference for detecting PCB defects in complex industrial environments.