Surface Defect Detection of Polarizer Based on Improved YOLOX-S Algorithm

doi:10.3778/j.issn.1002-8331.2209-0375

Abstract

Abstract: Polarizers are an essential part of liquid crystal displays, whose surface defects not only reduce the display quality of liquid crystal displays, but also cause the scrapping of the entire liquid crystal panel. Aiming at the problems that the surface defects of polarizers have large scale differences and various shapes, an improved YOLOX-S polarizer surface defect detection algorithm is proposed in this paper. The adaptive balanced feature pyramid (ABFP) module is proposed to sufficiently incorporate the multi-level features extracted by the backbone network, while increasing the detection branch through a single convolution to further enhance the multi-scale detection capability of the model. Then, an attention module CBAM is introduced in ABFP to focus on important features. In addition, the Mish activation function is used instead of the SiLU activation function while the CIoU loss function is adopted. The experimental results show that the improved algorithm achieves 92.97% and 55.16% of mAP50 and mAP50：95 on the polarizer surface defect dataset, which are 1.86 and 1.34 percentage points higher than that of YOLOX-S (FPN). Frames per second reaches 50, which basically meets the needs of industrial real-time detection.

Key words: polarizer surface, defect detection, YOLOX-S, adaptive balance feature pyramid, CIoU, Mish

摘要： 偏光片是液晶显示器的重要组成部分，其表面缺陷不仅会降低液晶显示器的显示质量，甚至可能造成整个液晶面板的报废。针对偏光片表面缺陷存在尺度差异大、形状变化多样的问题，提出一种改进YOLOX-S的偏光片表面缺陷检测算法。提出自适应平衡特征金字塔（ABFP）模块充分融合主干网提取的多级特征，并通过单个卷积增加检测分支，进一步增强模型的多尺度检测能力。在ABFP中引入注意力模块CBAM关注重要特征。采用CIoU损失函数的同时使用Mish激活函数替代SiLU激活函数。实验结果表明，改进的算法在偏光片表面缺陷数据集上的mAP50和mAP50：95分别达到92.97%和55.16%，相比YOLOX-S（FPN）提升了1.86和1.34个百分点，每秒检测帧数（FPS）达到50，基本满足工业实时检测的需求。

关键词: 偏光片表面, 缺陷检测, YOLOX-S, 自适应平衡特征金字塔, CIoU, Mish

CHEN Le, ZHOU Yongxia, ZU Jiazhen. Surface Defect Detection of Polarizer Based on Improved YOLOX-S Algorithm[J]. Computer Engineering and Applications, 2024, 60(2): 295-303.

陈乐, 周永霞, 祖佳贞. 改进YOLOX-S的偏光片表面缺陷检测算法[J]. 计算机工程与应用, 2024, 60(2): 295-303.

References

[1] YEN H N, SYU M J. Inspection of polarizer tiny bump defects using computer vision[C]//2015 IEEE International Conference on Consumer Electronics (ICCE), 2015: 525-527.
[2] 赖文威. 偏光片外观缺陷成像机理与检测技术研究[D]. 深圳: 深圳大学, 2017.
LAI W W. Research on the imaging mechanism and detection technology of polarizer appearance defects[D]. Shenzhen: Shenzhen University, 2017.
[3] LIU R, SUN Z, WANG A, et al. Real-time defect detection network for polarizer based on deep learning[J]. Journal of Intelligent Manufacturing, 2020, 31(8): 1813-1823.
[4] 李春霖, 谢刚, 王银, 等. 基于YOLOv3-Tiny-D算法的偏光片缺陷检测[J]. 计算机集成制造系统, 2022, 28(3): 787-797.
LI C L, XIE G, WANG Y, et al. Defect detection of polarizer based on YOLOV3-Tiny-D algorithm[J]. Computer Integrated Manufacturing Systems, 2022, 28(3): 787-797.
[5] 夏禹, 肖金球, 翁玉尚. 基于改进Faster-RCNN的偏光片表面缺陷检测[J]. 光学技术, 2021, 47(6): 695-702.
XIA Y, XIAO J Q, WENG Y S. Surface defect detection of polarizer based on improved Faster-RCNN[J]. Optical Technique, 2021, 47(6): 695-702.
[6] GE Z, LIU S, WANG F, et al. YOLOX: exceeding YOLO series in 2021[J]. arXiv:2107.08430, 2021.
[7] LIN T Y, DOLLAR P, GIRSHICK R, et al. Feature pyramid networks for object detection[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017: 2117-2125.
[8] LIU S, QI L, QIN H, et al. Path aggregation network for instance segmentation[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018: 8759-8768.
[9] PANG J, CHEN K, SHI J, et al. Libra R-CNN: towards balanced learning for object detection[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2019: 821-830.
[10] MISRA D. Mish: a self regularized non-monotonic neural activation function[J]. arXiv:1908.08681, 2019.
[11] 王颖颖. 基于改进深度残差网络的小目标检测与跟踪研究[D]. 北京: 北京化工大学, 2021.
WANG Y Y. Research on small target detection and tracking based on improved deep residual network[D]. Beijing: Beijing University of Chemical Technology, 2021.
[12] ZHANG H, CISSE M, DAUPHIN Y N, et al. mixup: beyond empirical risk minimization[J]. arXiv:1710.09412, 2017.
[13] REN S, HE K, GIRSHICK R, et al. Faster R-CNN: towards real-time object detection with region proposal networks[C]//28th International Conference on Neural Information Processing Systems, 2015: 91-99.
[14] LIU W, ANGUELOV D, ERHAN D, et al. SSD: single shot multibox detector[C]//European Conference on Computer Vision. Cham: Springer, 2016: 21-37.
[15] LIN T Y, GOYAL P, GIRSHICK R, et al. Focal loss for dense object detection[C]//Proceedings of the IEEE International Conference on Computer Vision, 2017: 2980-2988.
[16] REDMON J, FARHADI A. Yolov3: an incremental improvement[J]. arXiv:1804.02767, 2018.
[17] LAW H, DENG J. CornerNet: detecting objects as paired keypoints[C]//Proceedings of the European Conference on Computer Vision (ECCV), 2018: 734-750.
[18] TIAN Z, SHEN C, CHEN H, et al. FCOS: fully convolutional one-stage object detection[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision, 2019: 9627-9636.
[19] ZHANG S, CHI C, YAO Y, et al. Bridging the gap between anchor-based and anchor-free detection via adaptive training sample selection[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020: 9759-9768.
[20] ZHANG H, WANG Y, DAYOUB F, et al. VarifocalNet: an IoU-aware dense object detector[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021: 8514-8523.
[21] FENG C, ZHONG Y, GAO Y, et al. Tood: task-aligned one-stage object detection[C]//2021 IEEE/CVF International Conference on Computer Vision (ICCV), 2021: 3490-3499.
[22] DUAN K, BAI S, XIE L, et al. CenterNet++ for object detection[J]. arXiv:2204.08394, 2022.