Research On Surface Defect Detection Method Of LCD Screen Based On Deep Learning

In recent years,with the improvement and development of computer vision and deep learning,it has become a key technology in the new era of intelligent industry.However,there is still a problem that the efficiency and accuracy cannot be completely unified in product quality inspection and small defect object detection.In order to further improve the efficiency and accuracy of LCD screen defect detection,the accuracy of LCD screen defect detection is improved in moving images.And real-time is the main content of the article research.The traditional LCD defect detection method is manual visual recognition,which mainly relies on human experience,and the efficiency of the entire identification process is low.The current difficulties in surface defect detection: multiple influencing factors such as ambient light and noise,insufficient online detection algorithm capability and low real-time performance,how to simulate the information processing function of the human brain to build an intelligent machine vision system,and how to solve fuzzy features in accurate identification.However,there are also problems such as the inability to balance the real-time performance and accuracy of the model,and the high missed detection rate and false detection rate.To this end,this paper studies and explores from the following aspects:（1）Aiming at the problems such as the impact detection of ambient light encountered in defect detection at present,this paper has done a lot of work in the process of data acquisition and image preprocessing,and designed a data acquisition system to effectively avoid the influence of environmental factors on the data acquisition effect.,Aiming at the analysis of the defect types and characteristics of small objects on the LCD screen,avoiding the problem of model missed detection,and combining image enhancement through multi-marker connection,the object detection network model can improve the detection accuracy of small objects and other subtle defects,and improve the detection accuracy of multi-object and small objects.Identifying ability and reducing the proportion of model missed detections.（2）The advantages and disadvantages of the two-stage object detection algorithm and the single-stage object detection algorithm in the object detection algorithm are analyzed,aiming at the problems that the two-stage object detection algorithm has high accuracy but is difficult to detect small objects and has poor real-time performance.An improved Mask R-CNN model based on transfer learning is proposed to detect defective LCD screens.Aiming at the problem that the single-stage object detection algorithm has strong real-time performance but inaccurate detection of small object defects and low overall detection rate,a YOLOv5 model based on transfer learning and improvement is proposed to detect defective LCD screens in real time.The experimental results show that the improved Mask R-CNN has a detection accuracy of95.25% for defective LCD screens,and it takes 98 ms to detect each image,which cannot meet the requirements of real-time detection,while the improved YOLOv5 s has an average detection accuracy of 98.82% for various LCD screen defects,the detection of each image reaches 22 ms,which meets the requirements of real-time detection.（3）In view of the fact that the deep learning detection method for LCD screen defects should be applied to the actual detection environment,under the premise of ensuring accuracy and real-time performance,the detection rate and detection time of the improved YOLOv5 s detection model and the Mask R-CNN detection model were compared and analyzed.And use visual analysis to further illustrate the rationality of the improved defect detection method proposed in this paper.In measuring the model size,detection rate and detection time,the final choice is based on the improved YOLOv5 model to achieve real-time LCD defect detection.