Glass Panel Defect Detection System Based On Industrial Vision

The defect detection of glass panels is a difficult point in industrial inspection,and an important problem that needs to be overcome urgently in the field of quality inspection.However,due to the mirror surface characteristics of the glass,the detection is very difficult,and the detection is still mainly performed by the human eye at present.Although the current detection devices on the market can play a part of replacing manual labor,they still have problems such as low detection efficiency,poor detection accuracy,and unstable detection results.In order to solve the above-mentioned problems,this article uses a combination of area scan camera and line scan camera from the perspective of industrial vision to locate,classify,and estimate the defect size of the glass panel.And on this basis,a complete detection device is designed.The main content includes the following aspects:1.Use industrial vision solutions for defect detection processing.Choose a suitable sensor camera for shooting by considering the characteristics of the glass panel.Then compare the camera parameters and lighting schemes,find a suitable shooting method to obtain images,and provide a hardware basis for subsequent image processing.2.The structured light method is used to perform three-dimensional restoration of the images taken by the area array camera.First,analyze the structured light model,and establish a complete shooting model combined with camera imaging.Use the black and white grid to calibrate the system parameters to obtain the system parameters.Substituting the phase image of the photographed object into the structure model to obtain the three-dimensional point cloud data of the glass panel,so as to realize the restoration of the three-dimensional surface shape.Determine the location of the flaw according to the depth image,provide the range of the flaw for image processing,and facilitate the segmentation and extraction of the flaw.3.Use Yolo neural network to perform defect recognition processing on the image.Firstly,the defects of the sample data are segmented and extracted,and the defect sample data set is obtained.Then define different defect categories,and identify the image according to different defect characteristics.Then use the Yolo network to train the sample data to obtain a network model that meets the characteristics of the defect.Substituting the test data into the model can select defects.Realize the estimation of the flaw size according to the actual shooting parameters of the line scan camera.4.An embedded interactive application for defect detection is designed for the overall process.The image is displayed according to the shooting process,and the shooting effect can be observed in real time.At the same time,the power management module adopts the Kalman filter algorithm to evaluate the power of the system,so that the stability of the light source can be judged and the lighting intensity can be guaranteed.Finally,the image processing algorithm is integrated into the system module to realize the flow processing.The whole process can display the measurement results on the interactive interface,visually observe the size of the flaw and give the corresponding estimated data.In this project,a complete image shooting and processing system was designed to realize the defect detection,size measurement and display of the glass panel.Through actual tests,the measurement results are stable and have good adaptability to most glass defects.