Research On Defect Detection Method Of Mobile Phone Screen Based On Deep Learning And Machine Vision

With the continuous development of science,technology and industrial level,people's quality requirements for LCD screen products such as mobile phone screen are also increasing.In order to prod uce high-quality and high-resolution mobile phone screens,the quality inspection of the glass screen is an essential procedure in the production process,it includes judging whether the screen products have defects,locating defects,classifying and grading defects.Therefore,how to quickly and accurately realize the defect detection of mobile phone screen is an urgent problem to be solved in the production line of mobile phone screen.At present,the detection of mobile phone screen and other transparent glass products mainly relies on manual detection or traditional machine vision detection.Manual detection cannot guarantee the high efficiency and accuracy of defect detection due to subjective actors,low efficiency and easy fatigue,while the traditional machine vision detection method can meet the demand of real-time and high efficiency,but its feature extraction ability is limited,and the accuracy cannot meet the requirement of defect detection performance.Deep learning technology has the ability to automatically learn useful features,and it can extract features layer by layer to take the advantages of the advantages and make up the disadvantages of the disadvantages.In view of the above problems,this paper carried out the research on the de fect detection method of mobile phone screen based on deep learning combined with machine vision.Aiming at the problem of low detection rate caused by insufficient useful feature extraction in the defect detection process of mobile phone screen,a method based on deep convolutional neural network advantage optimization template matching is proposed to improve the information integrity and reliability of feature template.Through the optimized template matching method,the flaw detection is improved,and the recognition degree of the flaw is improved,and the automatic classification and recognition of the flaw is realized.The method of template matching of machine vision is simple in implementation,high in sensitivity and good in robustness.Combining the accurate and automatic feature extraction ability of the deep convolutional neural network,the outstanding feature abstraction and expression ability,it realizes the efficient and accurate detection of the screen defects of the mobile phone,and also completes the classification of defects while detecting flaws.The cumbersome process of separately classifying defects is avoided.Aiming at the problem that the defect classification of the same type but different degrees of defect feature information is not obvious,the difficulty level is difficult to divide and the accuracy is low.A method based on deep learning and feature visualization technology is proposed.The feature information of the defect extracted by the convolutional neural network is visually analyzed by deconvolution technique,and the data set of the defect is accurately marked according to the size of the characteristic saliency value after the visualization.Then the hierarchical model training is carried out by means of the transfer learning of convolutional neural network to realize the classification of defects.This scheme not only guarantees the reliability of sample labeling,but also effectively solves the problems of insufficient sample data sets and lengthy training time,thus ensuring the accuracy of classification.The results show that the method of defect detection of mobile phone screen based on deep learning combined with machine vision not only significantly improves the detection rate of defects,but also significantly improves the accuracy of classification and grade.