Key points are not available for this paper at this time.
In view of the low detection efficiency and high missed detection rate in the current printed circuit board (PCB), this paper proposes an improved YOLOv3 PCB surface defect detection method. This method is based on the YOLOv3 network model. The improvement of its network structure mainly includes: 1. Combine the batch normalization (BN, Batch Normalization) layer to the convolutional layer, improve the forward reasoning speed of the model, and reduce the model's PCB defects the training time of the dataset. 2. Aiming at the problem that the objective function and evaluation metric are not uniform in the YOLOv3 object detection algorithm, the GIoU performance metric and loss function are used to improve the detection effect of the model on small and medium targets of PCB defects. 3. Use the K-means++ clustering algorithm to optimize the K-means clustering algorithm, and determine the appropriate anchor boxes for the PCB defect dataset. 4. Multiscale training is used to enhance the robustness of the model for image detection with different resolutions. The experimental results show that mAP (Mean Average Precision) reaches 92.13%, and the detection rate is increased to 63f/s, which is improved compared to the YOLOv3 model, and has a better application prospect in PCB surface defect detection.
Zhuo et al. (Fri,) studied this question.