Research On Key Technology Of Intelligent Optical Inspection Of Computer Light Guide Plate

In today's information age,computers still occupy a large market share due to their excellent performance.The computer light guide plate(CLGP),as an important part of the backlight module in computer monitors,is responsible for converting a linear light source into a surface light source,and its quality directly impact on the effect of screen display.In the production process of CLGPs,due to the influence of factors such as process,environment and so on,defects such as bright spots,scratches,and dirt will inevitably appear.The existence of these defects will affect the use of related equipment,so it is particularly important to check the quality of the LGPs and eliminate inferior products.At present,most LGP manufacturers use manual inspection methods,which are inefficient,inaccurate,and unhealthy for employees.With the improvement of market demand,the LGP production line must improve the quality and speed.Therefore,a small number of LGP manufacturers have begun to introduce traditional machine vision inspection devices to replace manual inspection methods.However,due to the uneven distribution of the light guide points of the CLGP and the different sizes,shapes,and brightness of the LGP defects,it is still difficult for traditional machine vision methods to cope with these complex situations.Aiming at the current problems and difficulties in the field of LGP defect detection,this paper has developed an intelligent optical inspection system for the quality of CLGP.Finally,combined with a large number of related experiments,the reliability of the detection system in this paper is verified.The full text is expanded from the following aspects:(1)In view of the optical characteristics of the LGP and the inspection requirements of the production line,a hardware solution for the defect detection system of the CLGP was developed,including mechanical architecture,the machine vision device,and the electrical control device.The mechanical architecture is responsible for system support,LGP transmission,LGP sorting and other functions;The machine vision device is responsible for image acquisition and processing.According to the defect imaging characteristics of the LGP,the selection scheme of devices is finally determined: a line scan camera,a line scan lens,a multi-angle light source,an industrial control computer;Electrical control device is responsible for the control logic of the entire system.The overall process is as follows: the LGP enters the inspection equipment from the production line,and is transported through the conveyor belt.First,the posture of the LGP is adjusted through the limit device to ensure that it does not exceed the field of camera view.Then,the LGP will trigger the photoelectric sensor to generate a feed signal which is transmitted to the PLC.After receiving the signal,the PLC lights up the multi-angle light source and triggers the line scan camera for image acquisition.Finally,based on the detection results,the PLC controls the manipulator to perform the sorting of LGP.(2)According to the image characteristics of the CLGP,a two-stage multiscale residual attention network based on "segmentation + decision" is constructed for the CLGP defect detection.In the first stage,a segmentation subnet is constructed to complete the precise segmentation and location of defects by using the U-shaped structure and the designed multiscale residual attention unit(MRAU).In the second stage,a decision subnet is constructed.Guided by the features extracted by the segmentation subnet,it can accurately determine whether the LGP image is defective.The designed model only uses a small number of samples for training,which overcomes the limit of defective sample size.Through a series of experiments on the self-built Compute Light Guide Plate Surface Defect Detection(CLGPSDD)dataset,and the open source high-resolution electronic commutator defect dataset(KolektorSDD),it is proved that the method of this paper has advantages in detection accuracy and versatility.Peculiarly,the accuracy rate on the CLGPSDD dataset reached 99.839%,and the F1-Score indicator reached 99.845%.(3)Based on the MFC framework of the VS2017 platform,combined with the OpenCV image library and the PyTorch deep learning platform,a modular LGP defect detection visualization system was developed.