1.昆明理工大学机电工程学院,云南 昆明 650500
2.河南中烟工业有限责任公司,河南 郑州 450000
网络出版日期:2024-10-11,
移动端阅览
孔祥飞, 王森, 赵林, 等. 可在TFT-LCD面板中实现多背景视觉细微缺陷检测的YOLO-DSM方法[J/OL]. 中山大学学报(自然科学版)(中英文), 2024,1-9.
KONG Xiangfei, WANG Sen, ZHAO Lin, et al. YOLO-DSM method for detecting multi-background visual micro-defects in TFT-LCD panels[J/OL]. Acta Scientiarum Naturalium Universitatis Sunyatseni, 2024,1-9.
孔祥飞, 王森, 赵林, 等. 可在TFT-LCD面板中实现多背景视觉细微缺陷检测的YOLO-DSM方法[J/OL]. 中山大学学报(自然科学版)(中英文), 2024,1-9. DOI: 10.13471/j.cnki.acta.snus.ZR20240261.
KONG Xiangfei, WANG Sen, ZHAO Lin, et al. YOLO-DSM method for detecting multi-background visual micro-defects in TFT-LCD panels[J/OL]. Acta Scientiarum Naturalium Universitatis Sunyatseni, 2024,1-9. DOI: 10.13471/j.cnki.acta.snus.ZR20240261.
提出了一种基于YOLO-DSM深度学习图像检测模型。首先,在每个Dark模块后引入HMU模块,以提高TFT-LCD面板上目标缺陷的检测精度。将原始SPP替换为SSMA,使得网络更加关注背景低对比度目标。其次,引入DSM模块以帮助网络增强有用特征且抑制无用特征,增强语义信息的集成。最后,用ODConv模块替换原始网络的下采样卷积,细化局部特征映射,实现局部缺陷特征的充分提取。在自制的TFT-LCD缺陷数据集中,与当前较为先进的算法进行对比。结果表明,YOLO-DSM网络在mAP精度方面达到了97.40%,且FPS达到了77.42帧,可满足TFT-LCD缺陷任务检测要求。
A deep learning image detection model based on You Only Look Once-Double Spatial-Squeeze Module (YOLO-DSM) is proposed. First,the Hierarchical Mixed-scale Unit (HMU) module is introduced after each Dark module to improve the detection accuracy of target defects on TFT-LCD panels. The original Spatial Pyramid Pooling(SPP) is replaced with Simple Spatial Mlp Attention(SSMA) to enable the network to focus more on targets with low contrast against the background. Second, the Double Spatial-Squeeze Module(DSM) is introduced to help the network enhance useful features and suppress useless ones, thereby enhancing the integration of semantic information. Finally, the Omni-dimensional Dynamic Convolution(ODConv) module replaces the down-sampling convolution of the original network to refine local feature mapping and achieve full extraction of local defect features. In comparative experiments on a self-made TFT-LCD defect dataset, the YOLO-DSM network achieved an mAP accuracy of 97.40% and an FPS of 77.42 frames. This meets the requirements of TFT-LCD defect detection tasks.
视觉细微缺陷YOLO-DSM全维动态卷积SCSE注意力机制
visual micro-defectsYOLO-DSMomni-dimensional dynamic convolutionspatial and channel squeeze & excitation
BOCHKOVSKIY A, WANG C Y, LIAO H Y M, 2020. Yolov4: Optimal speed and accuracy of object detection[EB/OL]. arXiv:2004.10934.
CHANG Y C, CHANG K H, MENG H M, et al, 2022. A novel multicategory defect detection method based on the convolutional neural network method for TFT-LCD panels[J]. Math Probl Eng:6505372.
CHEN M, CHEN P, WANG S, et al, 2022. TFT‐LCD mura defect visual inspection method in multiple backgrounds[J]. J Soc Info Display, 30(11): 818-831.
CHEN M, CHEN S, WANG S, et al, 2023. Accurate segmentation of small targets for LCD defects using deep convolutional neural networks[J]. J Soc Info Disp, 31(1): 13-25.
CHEN P, CHEN M, WANG S, et al, 2023. Real-time defect detection of TFT-LCD displays using a lightweight network architecture[J]. J Intell Manuf,35(3): 1-16.
DALAL N, TRIGGS B, 2005. Histograms of oriented gradients for human detection[C]//IEEE Computer Society Conference on Computer Vision and Pattern Recognition(CVPR'05).
HOU Q, ZHOU D, FENG J,2021. Coordinate attention for efficient mobile network design[C]//IEEE/CVF Conference on Computer Vision and Pattern Recognition.
HUANG W, LU H, 2012. Defect classification of TFT-LCD with bag of visual words approach[C]//IEEE 6th International Conference on Information and Automation for Sustainability.
LI C, ZHOU A, YAO A, 2022. Omni-dimensional dynamic convolution[EB/OL]. arXiv:2209.07947.
LIN G, KONG L, LIU T, et al, 2022. An antagonistic training algorithm for TFT-LCD module mura defect detection[J]. Signal Process Image Commun, 107: 116791.
LIN T Y, GOYAL P, GIRSHICK R, et al, 2017. Focal loss for dense object detection[C]//IEEE International Conference on Computer Vision.
MAO J, XU G, HE L, et al, 2023. Attention-relation network for mobile phone screen defect classification via a few samples[J]. Digit Commun Netw,10(4):1113-1120.
NGO C, PARK Y J, JUNG J, et al, 2017. A new algorithm on the automatic TFT‐LCD mura defects inspection based on an effective background reconstruction[J]. J Soc Info Disp, 25(12): 737-752.
PANG Y, ZHAO X, XIANG T Z, et al,2022. Zoom in and out: A mixed-scale triplet network for camouflaged object detection[C]//IEEE/CVF Conference on Computer Vision and Pattern Recognition.
ROY A G, NAVAB N, WACHINGER C, 2018. Concurrent spatial and channel ‘squeeze & excitation’in fully convolutional networks[M]. Cham: Springer: 421-429.
SUN Y, LI X, XIAO J, 2019. A cascaded mura defect detection method based on mean shift and level set algorithm for active‐matrix OLED display panel[J]. J Soc Info Disp, 27(1):13-20.
WANG C Y, BOCHKOVSKIY A, LIAO H Y M, 2023. YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors[C]//IEEE/CVF Conference on Computer Vision and Pattern Recognition.
WANG L, HUANG X, ZHENG Z, 2023. Surface defect detection method for electronic panels based on attention mechanism and dual detection heads[J]. PloS One, 18(1): e0280363.
YANG H, SONG K, MEI S, et al, 2018. An accurate mura defect vision inspection method using outlier-prejudging-based image background construction and region-gradient-based level set[J]. IEEE Trans Autom Sci Eng,15(4): 1704-1721.
ZHU X, LYU S, WANG X, et al, 2021. TPH-YOLOv5: Improved YOLOv5 based on transformer prediction head for object detection on drone-captured scenari-os[C]// IEEE/CVF International Conference on Computer Vision.
0
浏览量
0
下载量
0
CSCD
关联资源
相关文章
相关作者
相关机构