基于动态视觉传感器的铝基盘片表面缺陷检测

马居坡; 陈周熠; 吴金建

doi:10.16383/j.aas.c240307

基于动态视觉传感器的铝基盘片表面缺陷检测

doi: 10.16383/j.aas.c240307

马居坡^{1, 2,},
陈周熠^1,,
吴金建^{1, 2,}

1.
西安电子科技大学人工智能学院西安 710071
2.
琶洲实验室(黄埔) 广州 510555

基金项目: 国家重点研发计划(2023YFA1008500), 陕西省自然科学基础研究计划(2024JC-YBQN-0627), 中央高校基本科研业务费专项资金(XJSJ23079, ZYTS24006)

详细信息

作者简介:
马居坡：西安电子科技大学人工智能学院讲师. 主要研究方向为仿生动态视觉处理, 视觉缺陷检测. E-mail: majupo@xidian.edu.cn

陈周熠：西安电子科技大学人工智能学院博士研究生. 主要研究方向为仿生动态视觉处理、视觉缺陷检测. E-mail: chenzhouyi@stu.xidian.edu.cn

吴金建：西安电子科技大学人工智能学院教授. 主要研究方向为高质量成像和图像智能处理. 本文通信作者. E-mail: jinjian.wu@mail.xidian.edu.cn

计量
- 文章访问数: 147
- HTML全文浏览量: 62
- 被引次数: 0
出版历程
- 收稿日期: 2024-06-03
- 录用日期: 2024-07-23
- 网络出版日期: 2024-08-28

Dynamic Vision Sensor based Defect Detection for the Surface of Aluminum Disk

MA Ju-Po^{1, 2
,},
CHEN Zhou-Yi^1
,,
Wu Jin-Jian^{1, 2
,}

1.
School of Artificial Intelligence, Xidian University, Xi＇an 710071
2.
Pazhou Lab (Huangpu), Guanzhou, 510555

Funds: Supported by National Key Research and Development Program of China(2023YFA1008500), Natural Science Basic Research Program of Shaanxi (2024JC-YBQN-0627), and the Fundamental Research Funds for the Central Universities (XJSJ23079, ZYTS24006)

More Information

Author Bio:
MA Ju-Po　Lecturer at the School of Artificial Intelligence, Xidian University. His research interests covers bionic dynamic vision processing and visual defect detection

CHEN Zhou-Yi　Ph.D. candidate at the School of Artificial Intelligence at Xidian University. His research interests covers bionic dynamic vision processing and visual defect detection

Wu Jin-Jian　Professor at the School of Artificial Intelligence, Xidian University. His research interests covers high-quality imaging and intelligent image processing. Corresponding author of this paper

摘要

摘要: 现有视觉缺陷检测技术通常基于传统电荷耦合器件(Charge-coupled device, CCD)或互补金属氧化物半导体(Complementary metal-oxide-semiconductor, CMOS)相机进行缺陷成像和后端检测算法开发. 然而, 现有技术存在成像速度慢、动态范围小、背景干扰大等问题, 难以实现对高反光产品表面弱小瑕疵的快速检测. 针对上述挑战, 创新性地提出了一套基于动态视觉传感器(Dynamic vision sensor, DVS)的缺陷检测新模式, 以实现对具有高反光特性的铝基盘片表面缺陷的高效检测. DVS是一种新型的仿生视觉传感器, 具有成像速度快、动态范围大、运动目标捕捉能力强等优势. 首先开展了面向铝基盘片高反光表面弱小瑕疵的DVS成像实验, 并分析总结了DVS缺陷成像的特性与优势. 随后, 构建了第一个基于DVS的缺陷检测数据集(Event-based defect detection dataset, EDD-10k), 包含划痕、点痕、污渍三类常见缺陷类型. 最后, 针对缺陷形态多变、纹理稀疏、噪声干扰等问题, 提出了一个基于时序不规则特征聚合框架的DVS缺陷检测算法(Temporal irregular feature aggregation framework for event-based defect detection, TIFF-EDD), 实现对缺陷目标的有效检测.
- 缺陷检测 /
- 动态视觉传感器 /
- 高反光表面 /
- 不规则特征提取 /
- 时序融合 /
- 事件相机
Abstract: Current visual defect detection technologies usually rely on conventional Charge-Coupled Device (CCD) or Complementary Metal-Oxide-Semiconductor (CMOS) cameras for defect imaging and the development of backend detection algorithms. However, these technologies encounter challenges such as slow imaging speed, limited dynamic range, and significant background interference, which hinder the rapid detection of minor defects on highly reflective surfaces. To address these challenges, we innovatively propose a new defect detection mode based on Dynamic Vision Sensors (DVS) to achieve efficient defect detection on the highly reflective surfaces of aluminum disks. DVS is a novel bio-inspired visual sensor with advantages such as fast imaging speed, high dynamic range, and excellent ability to capture moving objects. First, we conduct a series of imaging experiments on the surface defects of aluminum disks and analyze the characteristics and advantages of DVS on defect imaging. Then, we establish the first event-based defect detection dataset (EDD-10k), including three common defect types: scratch, point and stain. Finally, to address the issues such as varying defect shapes, sparse textures, and noise interference, we propose a temporal irregular feature aggregation framework for event-based defect detection (TIFF-EDD), and realize the effective detection of defect targets.
- Defect detection /
- dynamic vision sensor /
- highly reflective surface /
- irregular feature extraction /
- temporal fusion /
- event camera
注释:

1) 2¹ 1微米 = 10000埃

2) 3² 由于在事件可视化时采用了同样大小的时间窗口, 快速运动的目标会产生更多的事件. 因此, 图6(c)比图6(a)具有更粗的边缘或拖影.

3) 1³ 某些缺陷盘片表面可能同时具有多种缺陷, 因此表1中不同类别的总事件流数量相加之后大于116.

HTML全文

图 1 DVS缺陷成像机理说明

Fig. 1 Explanation of the mechanism of DVS-based defect imaging

下载: 全尺寸图片幻灯片

图 2 事件产生数量与光强变化大小的关系示意图. ((a) 当$\Delta t$时间内光强变化小时, 产生的事件数量少; (b) 当$\Delta t$时间内光强变化大时, 产生的事件数量也相应变多)

Fig. 2 Relationship diagram between the number of generated events and the magnitude of light change. ((a) When the change of light intensity in $\Delta t$ is small, fewer events are generated; (b) When the change of light intensity in $\Delta t$ is large, the number of generated events increases accordingly)

下载: 全尺寸图片幻灯片

图 3 缺陷检测实验平台示意图

Fig. 3 Diagram of the defect detection experimental platform

下载: 全尺寸图片幻灯片

图 4 弱小缺陷、环境倒影的成像效果对比. ((a) DVS成像效果; (b)传统相机成像效果). 红色矩形区域内为表面划痕缺陷, 深度约为0.2微米, DVS对划痕成像特征明显, 传统相机无法有效成像. 黄色圆形区域内为环境倒影, DVS能够有效避免倒影引起的事件干扰, 传统相机对环境倒影依然成像. (圆形盘片外围的物体为夹具)

Fig. 4 Comparison of imaging effects of small defects and environmental reflections. ((a) The imaging effects of DVS; (b) The imaging effects of traditional camera). In the red rectangular area, there is a surface scratch defect with a depth of approximately 0.2 micrometers. The DVS can clearly captures the scratch features, whereas the traditional camera cannot effectively image it. In the yellow circular area, there is an environmental reflection. The DVS can effectively avoid event interference caused by reflections, while the traditional camera still captures the environmental reflection. (The object surrounding the circular disk is clamp)

下载: 全尺寸图片幻灯片

图 5 大动态范围成像效果对比. ((a) DVS成像效果; (b)传统相机成像效果). 红色矩形区域内为缺陷. 在强光情况下, 传统相机由于曝光过度, 无法有效成像; DVS由于其大动态范围特性, 能够实现对缺陷的有效捕捉. (盘片内孔处的物体为夹具)

Fig. 5 Comparison of high dynamic range imaging effects. ((a) The imaging effects of DVS; (b) The imaging effects of traditional camera). There is a defect in the red rectangular area. Under strong lighting conditions, the traditional camera fails to capture it effectively due to overexposure, whereas the DVS, with its high dynamic range, is able to capture the defect effectively. (The object at the inner hole of the disk is clamp)

下载: 全尺寸图片幻灯片

图 6 在不同运动速度下的缺陷成像效果对比. 红色矩形框内为缺陷所在区域. ((a)低速情况下, DVS成像效果, 缺陷成像明显. (b)低速情况下, 传统相机成像效果, 缺陷成像微弱, 可轻微识别出缺陷目标(电子版放大观看以取得最佳视觉效果). (c)高速情况下, DVS成像效果, 缺陷成像特征明显; (d)高速情况下, 传统相机成像效果, 难以识别出缺陷目标) (盘片内孔处的物体为夹具)

Fig. 6 The comparison of imaging effects under different motion speeds. There is a defect in the red rectangular area. ((a)At low speeds, the DVS imaging performance is effective, with the defect being clearly captured; (b) At low speeds, the traditional camera＇s imaging performance is weak, with the defect being faintly visible and only slightly recognizable (Best viewed in electronic version). (c) At high speeds, the DVS imaging performance is effective, with the defect features clearly visible; (d) At high speeds, the traditional camera's imaging performance makes it difficult to identify the defect) (The object at the inner hole of the disk is clamp)

下载: 全尺寸图片幻灯片

图 7 不同缺陷类别的事件图像以及标注框的可视化结果. ((a)点痕; (b)划痕; (c)污渍)

Fig. 7 Visualization of event images and annotation boxes for different defect categories. ((a) Point; (b) Scratch; (c) Stain)

下载: 全尺寸图片幻灯片

图 8 数据集EDD-10k统计特性. ((a) ~ (c)分别为点痕、污渍、划痕缺陷目标边界框宽高比的分布; (d)为所有缺陷目标边界框的面积分布; (e)为所有事件流的持续时长分布直方图; (f)为缺陷标签在训练集和测试集中的数量分布)

Fig. 8 Statistical characteristics of the EDD-10k dataset. ((a) ~ (c) illustrate the distributions of height-to-width ratios of bounding boxes for point, stain and scratch; (d) illustrates the distribution of areas of all object bounding boxes; (e) illustrates the time duration of all event streams; (f) illustrates the distribution of defect labels in the training and testing sets)

下载: 全尺寸图片幻灯片

图 9 缺陷检测器的主要网络框架

Fig. 9 Main network framework of the defect detector

下载: 全尺寸图片幻灯片

图 10 基于决策级时序预测融合DPF的后处理模块工作流程图. 红色预测框预测目标为污渍, 橙色预测框预测目标为点痕. 由于橙色预测框与前后时刻同类别的预测框的交并比低, 因此被过滤掉. 红色预测框与前后时刻同类别的预测框的交并比大, 因此被保留

Fig. 10 The workflow diagram of the post-processing module based on decision-level prediction fusion DPF. The red prediction box identifies the target as a stain, while the orange prediction box identifies the target as a spot mark. Since the $ IoU$ of the orange prediction box with prediction boxes of the same category from previous and subsequent frames is low, it is filtered out. The red prediction box, however, has a high $ IoU$ with prediction boxes of the same category from previous and subsequent frames, so it is retained

下载: 全尺寸图片幻灯片

表 1 事件流在训练集和测试集中的数量分布

Table 1 Sample distribution in the training set and testing set

类别	事件流总数	训练集事件流数量	测试集事件流数量
点痕	35	27	8
划痕	39	30	9
污渍	44	34	10
合格	4	2	2

下载: 导出CSV

表 2 EDD-10k数据集中每一类缺陷标签的数量

Table 2 The number of labels for each defect category in EDD-10k

类别	事件流总数	训练集事件流数量	测试集事件流数量
点痕	2865	2381	484
划痕	2159	1781	378
污渍	4070	3060	1010
合格	9094	7222	1872

下载: 导出CSV

表 3 与其他算法在EDD-10k数据集上的对比实验

Table 3 Comparison experiments with other algorithms on the EDD-10k dataset

Method	mAP	AP@点痕	AP@划痕	AP@污渍
Faster-Rcnn^[41]	0.210	0.000	0.536	0.095
YoloV5	0.569	0.393	0.756	0.559
YoloV7^[42]	0.543	0.471	0.644	0.514
RDN^[47]	0.512	0.553	0.476	0.507
MEGA^[48]	0.401	0.356	0.509	0.349
YoloV^[49]	0.537	0.112	0.628	0.670
SSD-event^[40]	0.236	0.087	0.626	0.138
SODformer-event^[25]	0.394	0.363	0.161	0.495
TIFF-EDD	0.617	0.512	0.701	0.639

下载: 导出CSV

表 4 在EDD-10k数据集上的消融实验结果

Table 4 Ablation study results on the EDD-10k dataset

	Baseline	MIFE	CTAA	VFLoss	DPF	mAP
TIFF-B	$\checkmark$					0.577
TIFF-MIFE	$\checkmark$	$\checkmark$				0.590
TIFF-CTAA	$\checkmark$	$\checkmark$	$\checkmark$			0.607
TIFF-VF	$\checkmark$	$\checkmark$	$\checkmark$	$\checkmark$		0.612
TIFF-EDD	$\checkmark$	$\checkmark$	$\checkmark$	$\checkmark$	$\checkmark$	0.617

下载: 导出CSV

参考文献(50)

[1]	金侠挺, 王耀南, 张辉, 刘理, 钟杭, 贺振东. 基于贝叶斯CNN和注意力网络的钢轨表面缺陷检测系统. 自动化学报, 2019, 45(12): 2312−2327 Jin Xia-Ting, Wang Yao-Nan, Zhang Hui, Liu Li, Zhong Hang, He Zhen-Dong. DeepRail: Automatic visual detection system for railway surface defect using Bayesian CNN and attention network. Acta Automatica Sinica, 2019, 45(12): 2312−2327
[2]	陶显, 侯伟, 徐德. 基于深度学习的表面缺陷检测方法综述. 自动化学报, 2021, 47(5): 1017−1034 Tao Xian, Hou Wei, Xu De. A survey of surface defect detection methods based on deep learning. Acta Automatica Sinica, 2021, 47(5): 1017−1034
[3]	Chen Y Q, Pan J W, Lei J Y, Zeng D Y, Wu Z Z, Chen C S. EEE-Net: Efficient edge enhanced network for surface defect detection of glass. IEEE Transactions on Instrumentation and Measurement, 2023, 72: Article No. 5029013
[4]	Jiang W B, Liu M, Peng Y N, et al. HDCB-Net: A neural network with the hybrid dilated convolution for pixel-level crack detection on concrete bridges. IEEE Transactions on Industrial Informatics, 2021, 17(8): 5485−5494 doi: 10.1109/TII.2020.3033170
[5]	Zheng Y J, Zheng L X, Yu Z F, Shi B X, Tian Y H, Huang T J. High-speed image reconstruction through short-term plasticity for spiking cameras. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Nashville, USA: IEEE, 2021. 6354−6363
[6]	Xu X Y, Sun D Q, Pan J S, Zhang Y J, Pfister H, Yang M H. Learning to super-resolve blurry face and text images. In: Proceedings of the IEEE International Conference on Computer Vision (ICCV). Venice, Italy: IEEE, 2017. 251−260
[7]	李家宁, 田永鸿. 神经形态视觉传感器的研究进展及应用综述. 计算机学报, 2021, 44(6): 1258−1286 Li Jia-Ning, Tian Yong-Hong. Recent advances in neuromorphic vision sensors: A survey. Chinese Journal of Computers, 2021, 44(6): 1258−1286
[8]	Pan L Y, Hartley R, Scheerlinck C, Liu M M, Yu X, Dai Y C. High frame rate video reconstruction based on an event camera. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2022, 44(5): 2519−2533
[9]	Han J, Yang Y X, Duan P Q, Zhou C, Ma L, Xu C, et al. Hybrid high dynamic range imaging fusing neuromorphic and conventional images. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023, 45(7): 8553−8565 doi: 10.1109/TPAMI.2022.3231334
[10]	Hu Y H, Liu S C, Delbruck T. v2e: From video frames to realistic DVS events. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). Nashville, USA: IEEE, 2021. 1312−1321
[11]	冯维, 徐仕楠, 王恒辉, 熊芝, 王选择, 翟中生. 逐像素调制的高反光表面三维测量方法. 中国光学, 2022, 15(3): 488−497 doi: 10.37188/CO.2021-0220 Feng Wei, Xu Shi-Nan, Wang Heng-Hui, Xiong Zhi, Wang Xuan-Ze, Zhai Zhong-Sheng. Three-dimensional measurement method of highly reflective surface based on per-pixel modulation. Chinese Optics, 2022, 15(3): 488−497 doi: 10.37188/CO.2021-0220
[12]	王颖, 倪育博, 孟召宗, 高楠, 郭彤, 杨泽青, 等. 彩色高反光表面自适应编码条纹投影轮廓术. 光学学报, 2024, 44(7): Article No. 0712001 Wang Ying, Ni Yu-Bo, Meng Zhao-Zong, Gao Nan, Guo Tong, Yang Ze-Qing, et al. Adaptive coding fringe projection profilometry on color reflective surfaces. Acta Optica Sinica, 2024, 44(7): Article No. 0712001
[13]	Lichtsteiner P, Posch C, Delbruck T. A 128 x 128 120db 30mw asynchronous vision sensor that responds to relative intensity change. In: Proceedings of the IEEE International Solid State Circuits Conference-Digest of Technical Papers. San Francisco, USA: IEEE, 2006. 2060−2069
[14]	Chen S S, Guo M H. Live demonstration: CeleX-V: A 1M pixel multi-mode event-based sensor. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). Long Beach, USA: IEEE, 2019. 1682−1683
[15]	Brandli C, Muller L, Delbruck T. Real-time, high-speed video decompression using a frame- and event-based DAVIS sensor. In: Proceedings of the IEEE International Symposium on Circuits and Systems (ISCAS). Melbourne, Australia: IEEE, 2014. 686−689
[16]	Gallego G, Delbrück T, Orchard G, Bartolozzi C, Taba B, Censi A, et al. Event-based vision: A survey. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2022, 44(1): 154−180 doi: 10.1109/TPAMI.2020.3008413
[17]	Lagorce X, Orchard G, Galluppi F, Shi B E, Benosman R B. HOTS: A hierarchy of event-based time-surfaces for pattern recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(7): 1346−1359 doi: 10.1109/TPAMI.2016.2574707
[18]	Lazzaro J, Wawrzynek J, Mahowald M, Sivilotti M, Gillespie D. Silicon auditory processors as computer peripherals. In: Proceedings of the 5th International Conference on Neural Information Processing Systems. Denver, USA: ACM, 1992. 820−827
[19]	马艳阳, 叶梓豪, 刘坤华, 陈龙. 基于事件相机的定位与建图算法: 综述. 自动化学报, 2021, 47(7): 1484−1494 Ma Yan-Yang, Ye Zi-Hao, Liu Kun-Hua, Chen Long. Event-based visual localization and mapping algorithms: A survey. Acta Automatica Sinica, 2021, 47(7): 1484−1494
[20]	Li Y Z, Wang H L, Yuan S H, Liu M, Zhao D B, Guo Y W, et al. Myriad: Large multimodal model by applying vision experts for industrial anomaly detection. arXiv: 2310.19070, 2023. (查阅网上资料, 不确定文献类型及格式是否正确, 请确认)
[21]	Song K C, Yan Y H. Micro surface defect detection method for silicon steel strip based on saliency convex active contour model. Mathematical Problems in Engineering, 2013, 2013: Article No. 429094
[22]	Tabernik D, Šela S, Skvarč J, Skočaj D. Segmentation-based deep-learning approach for surface-defect detection. Journal of Intelligent Manufacturing, 2020, 31(3): 759−776 doi: 10.1007/s10845-019-01476-x
[23]	Bergmann P, Fauser M, Sattlegger D, Steger C. MVTec AD–A comprehensive real-world dataset for unsupervised anomaly detection. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Long Beach, USA: IEEE, 2019. 9584−9592
[24]	Wang J L, Xu C Q, Yang Z L, Zhang J, Li X O. Deformable convolutional networks for efficient mixed-type wafer defect pattern recognition. IEEE Transactions on Semiconductor Manufacturing, 2020, 33(4): 587−596 doi: 10.1109/TSM.2020.3020985
[25]	Li D Z, Tian Y H, Li J N. SODFormer: Streaming object detection with transformer using events and frames. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023, 45(11): 14020−14037 doi: 10.1109/TPAMI.2023.3298925
[26]	Nguyen A, Do T T, Caldwell D G. Real-time 6DOF pose relocalization for event cameras with stacked spatial LSTM networks. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). Long Beach, USA: IEEE, 2019. 1638−1645
[27]	Kim J, Bae J, Park G. N-ImageNet: Towards robust, fine-grained object recognition with event cameras. In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). Montreal, USA: IEEE, 2021. 2126−2136
[28]	Ge Z, Liu S T, Wang F, Li Z M, Sun J. YOLOX: Exceeding YOLO series in 2021. arXiv: 2107.08430, 2021. (查阅网上资料, 不确定文献类型及格式是否正确, 请确认)
[29]	Duan K W, Bai S, Xie L X, Qi H G, Huang Q M, Tian Q. CenterNet: Keypoint triplets for object detection. In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). Seoul, Korea (South): IEEE, 2019. 6568−6577
[30]	Xiong Y, Li Z, Chen Y, Wang F, Zhu X, Luo J, et al. Efficient deformable ConvNets: Rethinking dynamic and sparse operator for vision applications. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle, USA: IEEE, 2024. 5652−5661. (查阅网上资料, 未找到本条文献信息, 请确认)
[31]	Bochkovskiy A, Wang C Y, Liao H Y M. YOLOv4: Optimal speed and accuracy of object detection. arXiv: 2004.10934, 2020. (查阅网上资料, 不确定文献类型及格式是否正确, 请确认)
[32]	Liu S, Qi L, Qin H F, Shi J P, Jia J Y. Path aggregation network for instance segmentation. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Salt Lake City, USA: IEEE, 2018. 8759−8768
[33]	Wang C Y, Liao H Y M, Wu Y H, Chen P Y, Hsieh J W, Yeh I H. CSPNet: A new backbone that can enhance learning capability of CNN. In: Proceedings of the IEEE/CVF conference on Computer Vision and Pattern Recognition Workshops (CVPRW). Seattle, USA: IEEE, 2020. 1571−1580
[34]	He K M, Zhang X Y, Ren S Q, Sun J. Deep residual learning for image recognition. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas, USA: IEEE, 2016. 770−778
[35]	Lin T Y, Dollár P, Girshick R, He K M, Hariharan B, Belongie S. Feature pyramid networks for object detection. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Honolulu, USA: IEEE, 2017. 936−944
[36]	Liu B D, Xu C, Yang W, Yu H, Yu L. Motion robust high-speed light-weighted object detection with event camera. IEEE Transactions on Instrumentation and Measurement, 2023, 72: Article No. 5013113
[37]	Li J N, Li J, Zhu L, Xiang X J, Huang T J, Tian Y H. Asynchronous spatio-temporal memory network for continuous event-based object detection. IEEE Transactions on Image Processing, 2022, 31: 2975−2987 doi: 10.1109/TIP.2022.3162962
[38]	Wang X T, Chan K C K, Yu K, Dong C, Loy C C. EDVR: Video restoration with enhanced deformable convolutional networks. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). Long Beach, USA: IEEE, 2019. 1954−1963
[39]	Zhang H Y, Wang Y, Dayoub F, Sünderhauf N. VarifocalNet: An IoU-aware dense object detector. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Nashville, USA: IEEE, 2021. 8510−8519
[40]	Iacono M, Weber S, Glover A, Bartolozzi C. Towards event-driven object detection with off-the-shelf deep learning. In: Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). Madrid, Spain: IEEE, 2018. 1−9
[41]	Ren S Q, He K M, Girshick R, Sun J. Faster R-CNN: Towards real-time object detection with region proposal networks. In: Proceedings of the 28th International Conference on Neural Information Processing Systems. Montreal, Canada: ACM, 2015. 91−99
[42]	Wang C Y, Bochkovskiy A, Liao H Y M. YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Vancouver, Canada: IEEE, 2023. 7464−7475
[43]	Zhu X M, Wang S, Su J J, Liu F, Zeng L. High-speed and accurate cascade detection method for chip surface defects. IEEE Transactions on Instrumentation and Measurement, 2024, 73: Article No. 2506612
[44]	Yuan M H, Zhou Y B, Ren X Y, Zhi H, Zhang J, Chen H J. YOLO-HMC: An improved method for PCB surface defect detection. IEEE Transactions on Instrumentation and Measurement, 2024, 73: Article No. 2001611
[45]	Wang Y R, Song X K, Feng L L. MCI-GLA plug-in suitable for YOLO series models for transmission line insulator defect detection. IEEE Transactions on Instrumentation and Measurement, 2024, 73: Article No. 9002912
[46]	Zhu J, Pang Q W, Li S S. ADDet: An efficient multiscale perceptual enhancement network for aluminum defect detection. IEEE Transactions on Instrumentation and Measurement, 2024, 73: Article No. 5004714
[47]	Deng J J, Pan Y W, Yao T, Zhou W G, Li H Q, Mei T. Relation distillation networks for video object detection. In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV). Seoul, Korea (South): IEEE, 2019. 7022−7031
[48]	Chen Y H, Cao Y, Hu H, Wang L W. Memory enhanced global-local aggregation for video object detection. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Seattle, USA: IEEE, 2020. 10334−10343
[49]	Shi Y H, Wang N Y, Guo X J. YOLOV: Making still image object detectors great at video object detection. In: Proceedings of the 37th AAAI Conference on Artificial Intelligence. Washington, USA: AAAI, 2023. 2254−2262
[50]	Jeong J, Park H, Kwak N. Enhancement of SSD by concatenating feature maps for object detection. In: Proceedings of the British Machine Vision Conference. London, UK: BMVC, 2017.