Research On Visual Inspection System For The Quality Of In-car Light Guide Plate Based On Machine Learning

With the continuous development of Telematics,the demand of in-car navigation display equipment has increased significantly.As an important component of in-car navigation display,the quality of in-vehicle navigation light guide plate is directly related to the display.During the production,in-car navigation light guide plate will have various defects such as yellowing,bright spots,dark spots,scratches and crushing because of the factors such as material characteristics,plate design,processing temperature,equipment wear,etc.The detection of defects of in-car navigation guides is still at the stage of manual optometry.The optical characteristics of light guide plate show that its defects must be visible under strong light.Human eyes are easily damaged when exposed to strong light for a long time.And the design and manufacture of in-car navigation light guide plate are becoming more and more sophisticated,some defects have been so subtle that they exceed the artificial naked eye detection limit.Meantime,yellowing defects have a extremely low probability of occurrence,but they cannot be missed.The detection often produces waste of human resources.Therefore,this article developed a machine learning-based visual inspection system for quality inspection of in-car navigation light guide plate.It mainly designed the system of hard-ware and software.Large amounts of data were collected to verify the real-time,accuracy,and stability of the system.This article mainly contains the following aspects.(1)Design of visual inspection systemThe design of visual inspection system included the design of hardware and software.Hardware included mechanical structure,visual device,and electronic control system.Aiming to different inspection targets,the yellowing defect inspection station and the surface defect inspection station were designed respectively.Mechanical structure and selection of visual equipment were carried out for each station.Complete electrical control process was designed to ensure the stable drawing and operation of the system.Besides,this paper designed the algorithm of the whole system.(2)Design of detection algorithm for yellowing defect of in-car navigation light guide plateFor the yellowing defects of in-car navigation light guide plate,this paper proposed a yellowing defect detection algorithm based on color space and SVM.Firstly,for removing background,the yellowing image was preprocessed to extract the region of interest by using digital image processing technology.Then,a Gaussian directional derivative was designed to filter the image of the region of interest.The upper edges of the light guide plate were extracted by on the filtered image.Each plate was located by the edge coordinates.A feature vector was constructed by color average components of R,G,B,L,A on the image of each light guide plate.Finally,this paper used the color feature vector to train the SVM model,and designed experiments to verify the performance of the yellowing defect detection algorithm.The results showed that the algorithm can effectively and stably detect the yellowing defects of in-car navigation light guide plate.(3)Design of detection algorithm for surface defects of in-car navigation light guide plateAccording to the imaging characteristics of in-car navigation light guide plate defects under industrial cameras,surface defects are divided into point defects,line defects,and surface defects.After preprocessing the image to obtain the region of interest,according to the detection speed and accuracy requirements,a light guide plate surface defect detection algorithm combining lightweight and cascaded deep learning network was designed.Firstly,based on the characteristics of defect distribution of the light guide plate,a lightweight two-classification network was designed to cut up the suspected defect area rapidly.Secondly,the improved Res Net network was used to construct a multi-classification network.The lightweight network and the multi-classification network were cascaded and merged,and diversified features were extracted from the segmented suspected defect regions to achieve accurate defect classification.Then,defect region could be located and recognized by predicting images which were from fixed windows sliding on the completed light guide plate.Finally,lots of experiments were conducted based on the self-building data set.The experimental results showed that the average accuracy of the detection algorithm for light guide plate defects detection was 98.4%,and the single detection time was 1.95 s/pcs.The accuracy and real-time had meet the requirements of industrial detection.