Research On Defect Detection And Recognition System Of Light Guide Plate Based On Machine Vision

As the core component of backlight module,light guide plate is widely used in household appliances,automobile instruments and other industries.At present,the quality inspection of light guide plate is basically based on manual inspection.In recent years,machine vision technology has developed rapidly and has been widely used in the industrial field.This topic as the starting point,through research to achieve a high-speed and high accuracy of light guide plate defect detection and recognition software and hardware system,has important research value.In this paper,the large-size light guide plate of 12-17 inch is taken as the research object.Aiming at the existing problems and difficulties in the field of defect detection of light guide plate,a software and hardware system for defect detection and recognition of light guide plate is developed based on the requirements of comprehensive use.Finally,the reliability of the defect detection system in this paper is verified through experiments.The specific research contents are as follows:(1)According to the optical characteristics and defect characteristics of light guide plate,a visual scheme of light guide plate defect detection is designed,and the hardware selection scheme of machine vision system is determined: line-array camera,line-array camera adaptation lens,multi-angle line light source,and motion platform.Through the establishment of the system hardware platform to obtain the light guide plate image.(2)According to the actual production demand of the detection rate of 99% and 12s/PCS,research was conducted on the detection algorithm of light guide plate defects by taking the accuracy rate and real-time factors into consideration.In this paper,an algorithm based on traditional digital image processing is designed to detect the defects of light guide plate through median filtering,threshold segmentation,morphological processing and other operations.Then,aiming at the problems of low accuracy and poor adaptability of the traditional detection algorithm,the light guide plate defect detection algorithm based on deep learning is studied.Several classical deep learning algorithms are compared and studied respectively.Through the test results on the defect data set of the self-built light guide plate,the YOLOv5 s model with the highest accuracy is selected for improvement.(3)An improved YOLOv5 s network is proposed for defect detection of large size light guide plate.First,the light guide plate image was divided into different images,and then Transformer and Attention mechanism Coordinate Attention were integrated in the main part and feature fusion part,and the meta-Acon activation function was selected.Finally,a large number of experiments are carried out based on self-built data set LGPDD.Experimental results show that the average accuracy(m AP)of LGP defect detection algorithm can reach99.20%,and the FPS can reach 77.(4)Based on the above research results,the light guide plate defect detection and recognition system software was designed and implemented,and the system deployment and integration were successfully completed.After testing the light guide plate on the hot pressing line of a class of customers,The results show that the system can detect the defects of bright spot,scratch,foreign body,bump and dirt in the 17-inch light guide plate with 99% accuracy in 11s/ PCS,and meet the actual production demand of light guide plate.