Research On Defect Detection Method Of Light Guide Plate Based On Deep Learning

The light guide plate is a key component in the backlight module of the LCD screen,and its quality directly affects the quality effect of the screen.In the manufacturing process of light guide plate,due to the influence of factors such as raw material composition,equipment usage,processing technology and workers' operation,the surface will inevitably have processing defects such as dirt,spotting and scratching.Therefore,it is necessary to carry out defect detection on the light guide plate.At present,the defects on the surface of the light guide plate are mainly detected by naked eye observation,only a few manufacturers use the traditional image processing methods.The defects of the light guide plate are still extremely small under the image of the image captured by the high-resolution industrial camera,and the characteristics of different defects are different,and the light guide points of the entire light guide plate are densely distributed and uneven,which leads that the traditional image processing detection methods requires experienced visual experts to carry out a large number of feature extraction algorithm programming work and expensive code maintenance costs,with low accuracy and poor stability.Therefore,this paper proposes two defect detection methods based on deep learning.Both of these methods can learn and extract the characteristics of the light guide plate defects by training the neural network to avoid the complicated feature extraction algorithm programming.The specific research contents of the full text are as follows:(1)According to the appearance characteristics of the light guide plate and the accuracy requirements of the manufacturer for defect detection,a set of visual solutions for surface defect detection of the light guide plate is designed,which mainly includes the hardware selection of industrial cameras,lenses,light sources,image acquisition cards,and industrial computers.The light guide plate image is obtained through the overall platform construction.The traditional image processing method is used to preprocess the acquired image of light guide plate to extract the region of interest.(2)Using a deep learning-based classifier to achieve rough positioning detection of light guide plate defects.First,make a binary classification data set of a light guide plate slice image;second,use transfer learning to retrain the pre-trained classification network GoogLeNet-V1 on the data set;further,the trained binary classification model is evaluated by using a variety of evaluation indicators,and the evaluation results show that the binary classification model has a good classification effect on the test set images;finally,the binary classification model is used to conduct cyclic traversal defect detection on the entire image of light guide plate,and the defects of light guide plate are located in an area composed of one or more 256×256 pixel squares.Compared with traditional image detection methods,this method can effectively improve the accuracy and stability of defect detection.(3)Using deep learning-based image semantic segmentation technology to achieve pixel-level precise positioning detection of light guide plate defects.First,mark the defects in the light guide plate slice image to make a defect segmentation data set;second,use transfer learning to retrain the pre-trained PSPNet semantic segmentation network to the data set;then,Use a variety of evaluation indicators to evaluate the trained defect segmentation model,and the evaluation results show that the defect segmentation model has a good defect segmentation effect on the test set image;further,the segmentation model is used to carry out cyclic traversal defect detection on the whole image of light guide plate;Since the separate deep learning-based semantic segmentation defect detection method usually cannot meet the industrial practical application requirements,it is necessary to combine the simple machine vision method to make a second judgment and screening of all suspected defect regions detected by the deep learning-based semantic segmentation method.Compared with the traditional image detection method and the classifier defect detection method based on deep learning in this paper,this method improves the accuracy of defect detection by 3.2%and 1.9%respectively.