An Improved YOLOv11 Algorithm for Printed Circuit Board Defect Detection

Abstract

To address the detection of tiny defects in complex PCB backgrounds and satisfy the real-time requirements of industrial applications, this study proposes an improved YOLO v11s-based method for PCB defect detection. Data augmentation was employed to improve the distinguishability of defect textures and image edges. In addition, the ADown lightweight downsampling module, the MSCB fine-grained feature enhancement module, and the iAFF adaptive feature fusion module were introduced into YOLO v11s to enhance the model’s ability to extract features of tiny defects and perform multi-scale feature fusion. Experimental results demonstrate that the improved model achieves a precision of 96.8%, a recall of 92.4%, and an mAP@50 of 95.2%, representing increases of 2.2, 4.2, and 2.8 percentage points, respectively, compared with the baseline model. Meanwhile, the GFLOPs decrease from 21.3 to 20.5, indicating that the proposed method offers both superior detection accuracy and a clear lightweight advantage.