A DEFECT DETECTION SYSTEM FOR PCB MANUFACTURING SYSTEM BY APPLYING THE YOLOV4 ALGORITHM

Abstract

Defect detection is recognized to be the most integral criterion for the
printed circuit boards (PCBs) quality in industrial manufacturing. The traditional
PCB inspection methods have several disadvantages such as time-consuming,
labor-intensive, environmental clutter - susceptibility, and inaccurate detection
ability. This paper offers a deep learning method for PCB defect detection. This
method builds an improved neuron network based on the YOLOv4 algorithm. A
CSPDarknet53 is used with feature pyramid networks as the backbone for feature
extraction. Secondly, the spatial pyramid pooling layer and the path aggregation
network are utilized to predict better mimic errors on the PCB components.
Finally, the experimental results indicate a more reliable and efficient
performance compared to the existing works.