Research On Detection Algorithm Of PCB Image Surface Defect

In the background and process of today's information age,printed circuit board(PCB),as the carrier and medium of information technology,has a wide range of applications,from mobile phones,computers and other electronic products that we often contact and use in our daily life to military aircraft,satellites and other fields.Due to the higher requirements of PCB in industrial production and the high integration of electronic products,the production of PCB is more refined and the wiring structure is more complex,which leads to the greatly increased probability of PCB with defects.PCB board must ensure that the line connection,line distance,appearance and other indicators meet the requirements of industrial production standards.Any problem or defect in any link will affect the final quality of PCB and become a waste board that can not enter the market.The common defects of PCB are short circuit,open circuit and other physical defects;At the same time,there are dust,oxidation and other false defects.PCB with true defects is not allowed to enter the market,PCB with false defects can be put into use after cleaning.Due to the precise production requirements of PCB,traditional defect detection methods,such as manual visual inspection,are difficult to support the detailed,efficient and accurate detection standard of PCB in industrial assembly line.Therefore,in the case of meeting the efficiency,it is particularly important to produce qualified PCB boards and eliminate the PCB boards that do not meet the quality standards from the production line to avoid them entering the market.Therefore,in order to reduce the false detection rate of PCB,improve the recognition rate of PCB image defects and reduce the complexity of the algorithm,this paper proposes a PCB image surface defect detection algorithm,which realizes the efficient,fast and accurate detection of PCB.The following research areas have been summarised through the study and analysis of PCB image defect detection:1.In the aspect of PCB image preprocessing,firstly,the PCB image in the database is slanted by Hough transform,which lays the foundation for the subsequent filtering preprocessing of PCB image;secondly,use the morphological adaptive image enhancement algorithm to enhance PCB image,highlighting the edge information of the PCB image;finally,the PCB image denoising algorithm based on improved NLM is used to denoise and enhance the PCB image And compared with the traditional denoising algorithm.2.In the aspect of defect feature extraction of PCB image,firstly,used the Watershed Based PCB image segmentation algorithm to segment the defect and non defect regions of the image,which is convenient for feature extraction later;secondly,the connected component based extraction algorithm is used to roughly locate the defect of PCB image,and roughly extract the position of the defect;finally,the improved region growing method is used to segment the defect image PCB image defect detection and feature extraction.3.In the aspect of PCB image defect detection,aiming at the problems of low detection efficiency,high false detection rate and high missing detection rate,this paper constructs a PCB defect detection network.Firstly,Faster R-CNN and feature pyramid convolution network are combined to detect PCB defects.The 693 PCB images in the database are expanded by data enhancement technology,and the Faster R-CNN model is trained on the expanded data set.Secondly,in the defect feature extraction stage,vgg16 network is used to extract the defect feature of PCB image;A suitable anchor is designed,and multi-scale feature fusion technology is used to detect defects in PCB image.Finally,in order to remove the duplicate box and reduce the false detection rate,soft NMS algorithm is used in the subsequent steps of defect detection to improve the accuracy of PCB image defect detection.At present,the defect detection algorithm of PCB image has some shortcomings in defect recognition rate and algorithm robustness.Aiming at these shortcomings,this paper mainly studies the PCB image in the above three aspects.The experimental results of PCB images show that: the algorithm proposed in this paper has remarkable effect on PCB image denoising,feature extraction and defect detection,and can effectively identify PCB short circuit,open circuit,hole missing,dust,oxidation and other defects.The algorithm has strong robustness,can meet the requirements of PCB image defect detection,and has certain practicability.The efficiency of PCB image defect detection is also significantly improved High.