Machine vision-based cutting process for LCD glass defect detection system

In this research, the automatic optical detection system is developed for detecting sectional profile and surface of thin-film transistor liquid crystal display (TFT-LCD) panels after being treated through cutting process. Traditional image processing inspection relying on pre-determined thresholding cannot achieve ideal results in slight defects glass substrates. The proposed pre-processing process was integrated with deep learning technique to further enhance inconspicuous addition, photoelastic reflection lighting used highlight subtle low-contrast images When tested photodetector, uniformed effect achieved by combining concentrated light source inner coaxial lens line order test coarseness-related characteristics so as indicate defect while intensifying contrast background. features, it conducted separating rib mark features U-Net network model learning; result, 100% accuracy can be achieved. defect, Auto Encoder learn background picture retrieved from original linear regression As a next step, again predict result subtracting picture, then position highlighted; 98% acceleration, revising weighting data format. terms Model, reading time has been shortened 4.28 s; individual prediction 0.29 Encoder, 19.23 0.94 s. detection, circular polariscope developed. During theory employed projecting circularly polarized vertically onto panel produce interfering halo at deformed area surrounding denting resulting are referenced identify defect. By screening mean luminance value sliding window discrete value, 90% meantime, also pinpoint that over 3° average angle change seen normal