Research On Electronic Component Detection Algorithm Based On Deep Learning

In recent years,with the progress of the society and people living standard rise,people also more and more demand for electronic products,electronic products demand is increasing,has driven up demand for electronic products,electronic components,but also to the quality and reliability of electronic components put forward higher requirements,so the detection problem of electronics is becoming an increasingly prominent research topics.Detecting electronic components can be divided into discrete electronic components and printed circuit board electronics testing,discrete electronic components test is to check whether a single electronic component appearance defects,printed circuit board components inspection is to stay in the assembly of electronic products on board a function of some electronic components whether there is any appearance defect.The detection problems of discrete electronic components and printed circuit board components can be summarized as classification problems and target detection problems in computer vision.In the previous electronic component detection methods,the traditional image processing method is mostly used,and the deep learning method is not adopted.The main reason is that the amount of defect data is small,while the deep neural network training needs more data.Therefore,effectively expanding defect image data under the existing conditions has become a prominent research problem,and the second is to improve the accuracy of detection.Have greatly contributed to the success of this paper is that further study of the data in the discrete electronic components classification problem expansion problem and the printed circuit board target detection problem of the network design problem,including the classification of discrete electronic components testing and printed circuit board in the target detection,with SMT capacitors and electrolytic capacitor as the research object to carry out the study of related issues.Mainly made the following innovations:1.Establishment of electronic component database.The image database ofchip capacitor component and printed circuit board was established by visiting the factory,collecting and photographing.2.Ipix2 pix based on illumination mask is proposed.The relevant GAN models,such as GAN,CGAN and Pix2 pix,were studied in depth.In view of the shortcomings of the network model of Pix2 pix which only USES shape mask,the generation antagonism network model based on shape mask and light mask was proposed.3.The proposed Ipix2 pix network model is used to generate defect images of patch capacitance,which improves the diversity and authenticity of generated defect images,effectively expands the image data set of patch capacitance,and strongly supports the classification experiment of patch capacitance.4.Electrolytic capacitance detection of printed circuit board based on YOLOv3.Three network models,YOLOv1,YOLOv2 and YOLOv3,were studied,and EC-YOLOv3 was proposed on the basis of YOLOv3 for the detection of electrolytic capacitance of printed circuit boards.