PPW-YOLO: An Improved PCB Defect Detection Model Based on yolo

Accurate diagnosis of circuit board faults is critical to the manufacturing of electronic devices. YOLO (You Only Look Once), as an extremely influential algorithm in the field of object detection, is widely applied due to its fast and efficient detection performance. However, the complex and fine - grained nature of printed circuit boards images, with their dense circuitry and diverse defect types, poses challenges in terms of detailed feature extraction and accurate localization. In order to improve the quality and yield of printed circuit boards (PCBs) and achieve efficient and accurate detection of PCB defects, this paper proposes a PCB defect detection method based on improved YOLOv8. The lightweight model PP - LCNet is used in the backbone network part, whose depth separable convolution and other designs reduce the number of parameters and computation, adapt to the characteristics of the circuit board image, and efficiently extract the key features; meanwhile, the NWD (Normalised Wasserstein Distance) loss function is introduced to measure the difference between predicted frames and real frames comprehensively, and to enhance the model's sensitivity to the details of the faults. The improved model is applied to the circuit board image dataset for fault diagnosis experiments, and the results show that the average detection accuracy of the improved network reaches 96.3%, which is 1.2% higher compared to the original network. The improved network effectively increases the accuracy of PCB defect detection and can better adapt to the industrial inspection environment.