Research On New Type Display Defect Detection Technology Based On Deep Learning

Flat-panel displays are moving toward clearer,thinner,and more colorful trends,and their application areas are becoming more and more extensive.However,in the actual production process,due to the complicated manufacturing process and various other factors,various screen display defects will be introduced on the screen,which will seriously affect the product quality.Therefore,defect detection has become a key link in optimizing the process and controlling the quality of the product..The traditional detection method based on manual observation is very easy to cause problems such as false detection,missed detection,and unobjective evaluation standards.The traditional screen defect detection algorithm based on machine vision has reached a bottleneck,with complicated process design,low accuracy,and insufficient intelligence.The visual inspection method based on deep learning image processing technology has developed rapidly and has been continuously applied in the field of industrial inspection.This paper is oriented to the needs of screen inspection for active light-emitting display screens such as OLEDs.Starting from the difficulty of the detection of display screen defects,combined with deep learning methods,integrating the mainstream trend of small sample industrial defect detection,the design includes image acquisition and preprocessing systems.And the overall design of the main algorithm system.A hardware platform was built as an image acquisition module,and interference factors such as moiré were adjusted and suppressed.After the screen image acquisition is completed,a preprocessing module including target area extraction and background texture suppression operations is established to complete the preprocessing of the screen image.The main algorithm system of this article mainly includes two parts: data enhancement module and classification recognition module.Aiming at the current pain points such as insufficient number of defect samples and insufficient diversity of display screens,a data enhancement scheme based on the UM-CycleGAN model is designed,using UADD generator and multi-feature layer supervisory discriminator.The effectiveness and advantages of this model in data enhancement are verified by setting up comparative experiments.As the core part of this article,the classification and recognition module focuses on the detection difficulties of display screen defects.A detection classification scheme based on the TFPN-Faster R-CNN model is designed,and the triple feature pyramid network(TFPN)structure is used to optimize the feature extraction network.Subsequently,experiments verified the model’s good performance in defect detection,especially the accu racy of Mura defect detection.And through comparative experiments,the advantages of this experimental model on the original data set and the enhanced data set are verified,and the effectiveness of data enhancement and structural optimization on the improvement of model performance is verified.The research results of this paper have certain guiding significance for promoting the automatic detection and classification of display screen defects,and can be applied to the field of display screen defect detection.