Grand Challenges in Printed Circuit Board Defect Detection Based on Image Processing

Abstract

It is difficult to avoid defects in the production process of printed circuit boards. Traditional manual methods are not effective. In recent years, deep learning has been gradually applied to printed circuit board (PCB) detection, but the accuracy is still not high. This paper starts with PCB image defect detection, preprocesses the image through image graying, grayscale histogram equalization and Gaussian filtering, uses machine vision technology to realize global defect detection and location, uses deep learning method in feature extraction, and then proposes a multi-scale feature fusion method, which has high quality of feature extraction, and forms a Resnet101 fine-tuning network. The network is trained and verified on the PCB image data set through migration learning. Experiments show that the recognition rate of the improved Resnet101 fine-tuning neural network is significantly improved, and its average accuracy is also higher. Finally, defect target detection of PCB image is realized based on fast RCNN algorithm. By improving the Faster RCNN algorithm, the efficiency and accuracy of target detection are improved. Experiments show that the average accuracy and mAP value of the algorithm are also higher than others, which verifies the effectiveness and versatility of the algorithm and has high practical value.