Research On PCB Defect Detection Technology Based On Deep Learning

Printed circuit board is a key component of the electrical interconnection of electronic components.It can effectively reduce wiring and installation errors,thereby improving the level of intelligent management and manufacturing efficiency.In today’s electrified and automated society,it has become an indispensable and important component.Because of the imperfection of the existing manufacturing process and the few Operation Mistakes,the PCB board has many kinds of complicated defects,gradually unable to keep up with the technical progress brought about by the increase in production,system functional testing is too close to the end of the line,can not find defects in the early and timely rework to reduce losses.Therefore,this paper proposes a deep learning-based PCB defect detection technology,with YOLOv5 as the core algorithm,which can quickly detect six types of PCB defects,thus effectively improve the detection efficiency and reduce the cost,improve the detection effect.Using the PYQT-based visual interface,we can easily do the actual detection.The main contents of this paper are as follows:Firstly,first,the development and research status of PCB defect detection methods are discussed,and the unique advantages of deep learning in PCB defect detection are expounded,the advantages and disadvantages of different algorithms and their application prospects are discussed.To improve the accuracy of deep learning,we selected 693 PCB defect detection data sets provided by the open laboratory of Intelligent Robotics,Peking University.To enhance the resolution of the images,In this paper,the data set through brightness adjustment,clipping and other data enhancement operations to expand the data set to 10668,the brightness is also higher than the original,in order to solve the problem of image quality degradation caused by the decrease of pixel density,ESRGAN is introduced to enhance the super-resolution of the cropped image,using the original data set and the images sampled under four times as the training ESRGAN data set,the image quality is effectively improved at the same time,Gauss filter is selected to denoise the image in the comparison of various filtering effects.considering the characteristics of large defect size and small figure size of PCB,a two-stage target detection algorithm YOLOv5 with high precision is proposed to train and detect PCB,and the feature pyramid is constructed for feature fusion,a detection head for detecting small objects is added to form a four-scale detection so as to improve the detection effect of small objects ASFF(adaptive feature space fusion)is added to the original FPN+PANET structure of YOLOv5,which ensures that each space can be fused with different levels of feature information adaptively Global Attention Mechanism was added to the original network,and attention operation was applied in all three dimensions(channel,space width and space height),which enhanced the ability of model information extraction.in order to transform the low-level algorithm into an easy-to-operate system,PyQt is used to design a visual interface,which can be used for quick operation and real-time defect detection for picture and video.