Research on circuit board fault detection algorithm based on computer vision technology

Abstract

In the actual use of the existing circuit board fault detection methods, the phenomenon of missing and misdetection often occurs, and the error detection rate is high. The problems existing in the traditional method will not only increase the cost of circuit board fault detection, but also can not provide accurate data for circuit board fault maintenance. Therefore, this paper proposes a circuit board fault detection method FPN50 based on deep learning. In this method, YOLOV5 is used as the detection model algorithm, and Relu in the original network is replaced by Relu6, so that the weights can be mapped more evenly, and the weight information can be retained more, so as to achieve quantization error. Secondly, the PAN structure is added after the original FPN network, which can enhance the positioning capability at multiple scales. The average accuracy of the final test reached 98.5%. Then the experimental results were verified with Shufflenetv2, Efficient net and Resnet50 detection models, and the average accuracy was 84.2%, 97.5% and 96.8%, respectively. The experimental results show that the FPN50 algorithm proposed in this paper has the highest detection accuracy and speed among all the comparison algorithms, and is more suitable for the detection requirements of this study.