Research On Visual Defect Detection Of Switching Electronic Components Based On Machine Learning

Switching electronic components are important constituent elements of electronic circuits.Due to the large variety of switching electronic components and diverse production processes,electrical parameter detection and manual detection methods are currently mainly used for detecting Switching electronic components defects.However,the electrical parameter detection method can only detect the finished product,the realtime performance is poor,and the manual visual detection method is inefficient.Longterm work.It may cause false detection and missed detection,which can't meet the needs of industrial production.Therefore,it is necessary to find a detection method with strong real-time performance,high efficiency and high detection accuracy.Aiming at some of the main process defects in the production process of switch electronic components,this paper studies the defect detection methods of switch electronic components based on display feature extraction,based on CNN,and based on two-stream convolutional neural network.Integrated detection network framework for switch electronic components based on two-stream neural network.The main research contents include:Aiming at the four major production defects of switch electronic components such as winding defects,welding defects,glue injection defects,and pin defects,a picture collection scheme of the main process defects of the switch electronic components was designed,and four defect data sets were produced through data enhancement and data standardization.In the welding defects and pin defects data sets,the same product corresponds to two pictures collected under different lighting conditions.The defect detection method of switch electronic components based on explicit feature extraction is studied.For different expressions of defect features in each data set,the defect features in the defect data set are extracted through algorithms such as image segmentation,image filtering,and morphological operations,and determines whether the product is good by comparing the extracted feature values with the standard values of good products.Defect detection method of switch electronic components based on convolutional neural network is studied.Using the classic convolutional neural network structure such as VGGNet,Inception module,residual block,etc.,a defect detection network for switch electronic components was built.The highest classification accuracy rate on the four test set of four data sets is 99.86%?99.76%?99.66%?99.51%.The defect detection method of switch electronic components based on two-stream convolutional neural network is studied.The main body of the network is divided into two parts.Two networks are used to process two pictures corresponding to the same product in the data set.two networks are fused at the back end of the network.The fusion method and fusion location of the two branch networks are discussed,and the classification accuracy of the welding and pin fouling data sets is tested.The classification accuracy of the highest is 99.88%and 99.56%on the two data sets.Finally,a network structure that performs well on all data sets is selected as branch network,and an integrated detection network framework based on the two-stream convolutional neural network is constructed and trained.A classification accuracy rate of 99.52%is obtained on the test set.