PCB surface defect detection method and device based on YOLOv3 algorithm

Abstract

The invention discloses a PCB surface defect detection method and device based on a YOLOv3 algorithm, and the method comprises the steps of building a network structure of the YOLOv3 algorithm, wherein a batch normalization layer and a convolution layer of the network structure are combined; determining a network loss function and a performance evaluation index of a YOLOv3 algorithm according to the GIoU; determining an anchor frame of the PCB surface defect sample data set according to a K-means + + clustering algorithm; establishing a PCB surface defect detection model based on a YOLOv3 algorithm; performing multi-scale training on the PCB surface defect detection model based on the YOLOv3 algorithm according to the PCB surface defect sample data set; and inputting the image data of the PCB to be detected into the trained PCB surface defect detection model based on the YOLOv3 algorithm, and outputting the position information of the surface defects of the PCB to be detected. According to the invention, efficient and high-precision detection of PCB surface defect detection is realized.

Description

technical field [0001] The invention relates to the technical field of target data detection, in particular to a PCB surface defect detection method and device based on the YOLOv3 algorithm. Background technique [0002] At present, the surface defect detection of printed circuit board (PCB) mainly faces the following challenges and difficulties: first, there are many types of PCB supply in the market, and the PCB design rules are different; second, the types of PCB defects and functional design are complex, Diverse; Third, the PCB defect detection industry lacks a large number of PCB defect samples, resulting in unbalanced data used in traditional methods. PCB surface defect detection mainly includes two parts: welding seam defect detection and component detection. Traditional manual visual inspection methods and machine vision inspection methods based on image processing have problems such as low precision, insufficient generalization ability, and poor robustness, and it ...

AI Technical Summary

Problems solved by technology

Traditional object detection methods also have inherent shortcomings: (1) The area selection algorithm based on sliding windows has low correlation, high time complexity, and a large number of redundant windows; (2) The artificial design function relies on a large amount of prior knowledge, subjectivity strong, and the detection process implemented in three steps is complex and cumbersome; (3) The traditional target detection method is slow and cannot meet the real-time detection requirements of industrial production

However, due to the local function of CNN, the existing deep learning algorithms are difficult to meet the requirements of PCB surface defect detection efficiency and accuracy

Images