Abstract: The production and application of Printed Circuit Boards (PCBs) continues to grow in the global industry and has become a core component of various electronic devices. Automatic defect detection of printed circuit board images is a challenging task due to the problem of small defect scales and the need for inspection models to be lightly embedded in portable devices. In order to meet the growing demand for high-quality printed circuit board products in smart manufacturing and usage, an improved YOLOv8-based defect detection method for printed circuit boards is proposed. First, a lightweight network MobileViT is used as the backbone network to reduce the model size and computation. Second, triplet attention module is introduced to enhance the ability of capturing features between different dimensions in the tensor. Finally, the bounding box loss function is replaced with L_\mathrmM\mathrmP\mathrmD\mathrmI\mathrmo\mathrmU to directly minimize the upper-left and lower-right point distances between the predicted box and the actual labeled box. Experiments show that the improved detection model can ensure high accuracy of small-size defect detection while having a very small number of parameters, and the reduction rate of the number of parameters of the model is 89.38%, which satisfies the applications of lightweight embedded portable inspection equipment and scenarios with limited computer resources, and confirms a good application prospect in the field of printed circuit board defect detection.