Research On Online Intelligent Visual Defect Inspection System For Mobile Phone Light Guide Plate

Nowadays,with the rapid popularization of Internet of Things(IOT)technology,smartphones and tablet PCs and other terminal devices have gradually become an important means for people to obtain information,and the screen,as a medium for displaying information,is in increasing demand.As an important component of the smartphone backlight module,the quality of the mobile phone light guide plate(MPLGP)is particularly important,which directly affects the display effect of the smartphone screen.In the production process of the MPLGP,due to the wear of the machining mold,the impure injection material,the production process and other factors,the light guide plate produced on the product line will inevitably have defects such as bright spots,line scratches,and black spots.At present,the quality inspection methods of MPLGPs by domestic LGP enterprises are still at the stage of manual optometry.Optometrists need to conduct product testing in a professional optometry environment.As a result of long-term work in an optometry environment with strong light,optometrists are prone to form eye fatigue,the human eye will suffer great damage,endangering their own physical and mental health.On the other hand,long-term and high-intensity optometry work also reduces the efficiency of optometrists and affects the quality of detection.As product inspection based on machine vision gradually replaces manual work and taking into account the current status of manual optometry for MPLGPs,the defect detection requirements of mobile light guide plate products are analyzed,and a set of defect online intelligent visual inspection system is developed.Finally,a large number of experimental analyses have been carried out to verify the reliability of the proposed algorithm.The full text mainly studies the following aspects.(1)By analyzing the imaging characteristics and discharge characteristics of the MPLGP,the mechanical design scheme and the visual imaging scheme of the MPLGP defect detection system are determined.Since the MPLGP production line is characterized by one mold and two holes,that is,two pieces of LGP products are produced from the production line at the same time.Meanwhile,it is necessary to ensure that the LGP cannot be damaged during the inspection process.The robot sucker is used to pick up the two products to the inspection machine one after another,and then they are inspected in turn with the movement of the conveyor belt.When the photoelectric sensor under the conveyor belt detects the moving LGP,the sensor sends a trigger signal to the line scan camera to perform image acquisition.According to the defect imaging effect,a set of image acquisition device is designed,which mainly includes 4K line scan camera,industrial lens,multi-angle light source and other equipment.(2)Design of recognition algorithm for module number of mobile phone light guide plateSince the LGP is tilted at a variable angle when it enters the inspection machine,and the actual center line scan lens is not exactly parallel to the plane where the LGP is located,this causes the captured image of the LGP to be tilted at an angle.If the module number is recognized directly,it is easy to cause the wrong recognition of the module number.In this paper,firstly,the tilt correction of the LGP image is performed.After the boundary of the MPLGP is obtained,the least square method is used to fit the boundary to a straight line.Finally,the projection transformation is used to perform tilt correction on the LGP image.On the basis of the tilt correction,the YOLOv4 object detection algorithm is used to identify the LGP module number.(3)Design of defect detection algorithm for mobile phone light guide plateFor the common defects such as bright spots,line scratches,and crush marks in MPLGPs,this paper proposed an end-to-end MPLGP defect detection network based on multi-task learning,which is mainly divided into four parts: encoder,multi-scale feature fusion,segmentation head and classification head.The encoder part is used to extract features at four different scales,while the multi-scale feature fusion part enables the interaction of feature information between multiple scales,and the multi-scale features after feature fusion are shared by the segmentation head and the classification head.The segmentation head is used to accomplish pixel-wise pinpointing of defects in images,while the classification head combines the multi-scale features and the output mask of the segmentation head for accurate judgments.The defect segmentation task and the classification task are learned in parallel to reduce the mutual influence between both tasks.Based on this,a self-built MPLGP defect detection dataset is constructed,and extensive experiments were conducted on the proposed multi-task learning-based defect detection network on the selfbuilt defect detection dataset and KoletorSDD.The experimental results showed that the algorithm has high detection accuracy for defects such as bright spots and line scratches in MPLGPs,and its F1-score reaches 99.71%,and the F1-score reaches 96.77% on KoletorSDD,which verified the accuracy and generality of the proposed algorithm.(4)Visual inspection system for mobile phone light guide plateIn this paper,the intelligent inspection system of MPLGP is developed based on Delsa camera library and OpenCV library together,and the visualization platform is built on VS2017 development platform using MFC framework,and the construction of the intelligent visual inspection visualization system for MPLGP is completed on Windows 10 operating system.