A Real-Time PCB Defect Detection Framework on Raspberry Pi Based on YOLOv5n

Abstract

PCB defect detection requires both high accuracy and real-time performance, while small defect sizes and complex backgrounds pose significant challenges for deployment on edge devices. This paper proposes a real-time PCB defect detection method based on the YOLOv5n model and constructs an edge-side detection system on Raspberry Pi 5. To enhance small defect detection capability, dataset integration, image slicing, and data augmentation strategies are employed. The trained model is deployed using the NCNN framework for lightweight inference. Experimental results demonstrate that the proposed method achieves 97.0% mAP50 on the test set and maintains a stable real-time detection speed of approximately 9 FPS on Raspberry Pi 5, showing strong performance on small-scale and sparse defects. The proposed approach provides a feasible solution for low-cost PCB defect detection on edge devices.