Research On Liquid Crystal Glass Defect Detection Technology Based On Convolutional Neural Networ

With the continuous development of the new generation of information technology,liquid crystal display devices have entered various industries and major fields.As the core component of liquid crystal display devices,liquid crystal glass has become a hot research direction for current display devices.The liquid crystal glass used in display devices has the characteristics of large size and thin thickness,so defects are prone to occur during the preparation process.Currently,many scholars have focused their research on the preparation process of liquid crystal glass,but relatively little research has been conducted on the surface defect detection technology of liquid crystal glass.In order to improve the yield and competitiveness of products entering the market,it is very important and necessary to conduct defect detection of liquid crystal glass in advance.Currently,the mainstream detection methods in this field are still manual or semi manual detection.The slow speed of manual detection and the fatigue caused by long-time work can reduce the detection accuracy.In order to improve the detection speed and accuracy,this paper proposes a defect detection method for liquid crystal glass based on convolutional neural networks.The main work is as follows:(1)Based on the defective liquid crystal glass collected from the production site,the data set required for the detection model was established,and the common defect types were divided into five categories: inclusions,bubbles,tin spots,nodules,and cracks.Manual labeling provided a reliable data set for the defect detection model.(2)Based on the Faster-RCNN two-stage target detection model,research was carried out on liquid crystal glass defect detection.The concept structure was introduced to improve the convolution layer in the feature extraction network VGG.The concept module reduced the amount of parameters for model learning,and through convolution of convolution kernels with different sizes,it was spliced and fused,making the features of the predicted feature layer richer,improving the detection accuracy and speed of the model.(3)In order to further improve the accuracy and speed of the detection model,this paper proposes a liquid crystal glass defect detection model based on an improved SSD.Res Net50 is introduced as a backbone network to increase the depth of the feature extraction network.The introduction of a feature pyramid structure,PANet,enables cross channel feature fusion on multiple prediction feature layers,and bidirectional fusion of low-level and high-level features,The channel attention SE module was inserted into the Res Net residual module to balance the weight of each channel feature extraction.Through experiments,it was found that the improved SSD model significantly improved the detection accuracy,especially the ability to detect small targets.(4)This article designs a visual liquid crystal glass defect detection system,which supports both automatic and manual detection modes,and supports continuous updates and upgrades of the system,making it convenient for users to use.