CC BY
5For citation: Tang Li, Shi Yishan, Xiang Xianwu, Mei Shunqi. An Examination of Algorithms for Target Detection and Their Application in Fabric Defect Detection Situations // Grand Altai Research & Education — Issue 2 (20)'2023 (DOI: 10.25712/ASTU.2410-485X.2023.02) — EDN: https://elibrary.ru/rsxsqc
UDK 677.014
An examination of algorithms for target detection
A
AND THEIR APPLICATION IN FABRIC DEFECT DETECTION SITUATIONS
Tang Li1, Shi Yishan1, Xiang Xianwu1, Mei Shunqi1
1 Hubei Digital Textile Equipment Key Laboratory, Wuhan Textile University, Wuhan, 430073, China E-mail: [email protected], [email protected]
Abstracts: Fabric defect detection is a significant area of research under the textile industry's growing trend toward automation and intelligence. Deep learning-based target identification algorithms have been applied extensively in the field of fabric detection in recent years, which has tremendously aided in the advancement of intelligence in the textile sector. The following factors are taken into consideration when analyzing the research state of YOLO series algorithms in the field of fabric flaw identification. It begins by summarizing the target detection development trend. Next, it summarizes and examines the structure and function of the YOLO family of algorithms. Finally, it talks about the use of YOLO algorithms and their derivatives in the field of fabric flaw identification and inspection. Lastly, it considers the issues and potential paths for target detection development in the future.
Keywords: target detection, YOLO, fabric defect detection
m^1, ^i^1, fift^1, mm1
1 430073
E-mail: [email protected], [email protected]
m YOLO
* This paper was supported by the National and Hubei Provincial High-end Textile Equipment Intellectual Intelligence
Base Programme (111HTE2022002, HWZ201819). ** (111HTE2022002, HWZ201819).
тЖ, YOLO ЖШШ^ЗДШ; ШФ^ШШТ YOLO
М^Й.
ШП: ЭШШ; YOLO; ШШ&ШШШ
0
ттш^мтшттжшшт^^-, ш&штшт
шшлмта, ишш. ЙХ^Х^ШЙЗД^ШШШШ, Й ЩШйШШЩ.
ш. шшшйш&штххштш, тт^ тшм,
ШМ, [1,2].
фй^вд, ЙШШШ^Й, шшш^т^, ^шймм&тжш ^шжш.
1
2014 2014 ^тшшмшэышшшшт, 2014
^^^Ш^ЗОДЭШЭДМЭД [3]. 1.
Ш 1 ШЫШШШЖМШ^М
Fig. 1. Evolution of target detection algorithms
1.1
ШШ&^ЭД, Ш.ШШШШ Viola Jones Detector [4].
ШШР Ы.
Ш, RÄ, (К^ШХШ^), HOG Detector [5]
DPM Detector [6] (Hard negative mining), &
(Bounding box regression) (Context priming)
m imx^mmummmmx^xmmm,
(Convolutional
Neural Network, CNN) (Deep Neural Network, DNN).
Convolutional nmEx^mmm, ^w^xrnx.
1.2
: (anchor-
based) (anchor-free). (One Stage)
ft^&g^&tl (Tow Stage). (№ 2):
UM, (region proposal, » RP,
(Spatial Pyramid Pooling Layer, SPP) ^T^,
R-CNN, Fast R-CNN, Faster R-CNN.
тктш
(Ш 3): шшя жШ^ФЕЙ, ът. штжшшштяяш, тшжшж'\\ m
ШШЖ&, Ш^тШШШ^т ШЖШЯ SSD, YOLO Ш.
Fig. 3. General flowchart for one-stage target detection
штштжшшшм: ЮТМйй^Ъ», нвдплт^м&ш, шпшш
тт.
.
Anchor free ^МХ 2015 ^ CVPR fä DenseBox.
,) anchor, ЖШШ& ^ВШИ. ЭД1Ш, Anchor free
тш^, жш^ШШМФ, ШШШЯЖ,
yolo
2.1 YOLO
YOLO. 4. *
^ YOLO ^ij, . YOLO
(1) (Backbone),
^m. (2) (Neck),
^ (Head),
Fig. 4. General architecture of YOLO algorithm 2.2 YOLO MMM&RR&M
yolovi m^rnwm^, Mmm 24 ^«rM^P 4 ^s^mfcr^w
YOLOv2 [6] (1) Batch normalization, ^
^IM^BK, (2) S^ anchor ISfgftRfl. #
YOLOvl + + ! + S^ anchor
^^mrnsuftRrninrn, rn&WMBfc, it«BJ\ fe&BM. (3) k-
means M^teUM anchor. (4) (5)
YOLOv3 [8] Backbone. ^ ResNet IM^i^, YOLOv3 i£if
7BMl Backbone, 19 ^7 53, &RMWT
2 ift^M, 3x3, mmrnwrn^, SMMP
YOLOv3 mfflHfflK^l^«,
G^ffl^NR-imrnftRrn. ft^ffl spp »^wm^i^«^ ^
^M^ifW^MM, ffl CIoU Loss DloU Loss, Bil^S^ft
R^mtiftR^is^.
YOLOv4[9] f^ffl CSPNet IM^tl CSP v3 l Darknet53,
3IA PANet PAN
^ (Eliminate grid sensitivity), (Mosaic data augmentation),
(Optimizered anchors)
YOLOv5 [10] Ä-^fättl&^fä, ü&^fäH&Ätt, ^MÄM'fc
mm anchor «m ^xmm
anchor ^ÄABt, anchor, MMEffi^
^mm^. ^ Warmup and Cosine LR scheduler, EMA
(Exponential Moving Average) INt YOLOv5 iffl PyTorch
YOLOv6 [11]
a7 Rep-PAN m YOLOX mg^fö
ÄfsE^MsE^M,
iffl SloU Loss,
YOLOv7 [12] fltftN^I^, i£if ELAN
(Efficient Layer Aggregation Network). + . ÄS®
yolov7
YOLOv8 [13] TOMH^SÄ«, SOTA
»Jm
3.1 YOLOv2
Zhang [14] m YOLOv2 ^ffl^^KM^«^, Älättt YOLO9000, YOLO-VOC, Tiny-YOLO !#iM,
№ YOLOv2 Xt YOLOv1 Äfr7&Ä, 7 YOLOv2 R
3.2 YOLOv3
fi« [15] &Ä Tiny-YOLOv3 I^fMHf
1x1 mmmm^Mmmmmm^,
&ÄH& mAP 3.1%, fH^MÄ^m
3.3 YOLOv4
[16] rnsx yolov4 imm&Mfflx^ffliximmm&wrn. it m yolov4 ntmtm^guimm^x^, ##tt^MlRrn sm
yolov4
ZRMimrn®. s ufomi&g&rnBM. PANet ^mmm, ^m^
im K-means++ M«S anchor R—. ^Wt, YOLOv4 Wtel
precision b№ YOLOv4 Wtei^^ 2.00%, mAP 2.72%.
3.4 YOLOv5
Zhou [17] YOLOv5 Backbone
N0ti£if7 ResDCN m
Focal EloU M&am,
3.5 YOLOv7
[18] ^»MlR®, » YOLOv7
DR-SPD, DRes
«nt^n spd
3.6 YOLO
s^m yolo ^w^wnMiM^^^tt^lfm&Rfi, &
4 ^
YOLO ^Wtel^MM
SM YOLO
lfflft, {H^Mil^^fflm^iM. ^M YOLO
[1] Henry Y., Grantham K.P., Nelson H.Y. Automated fabric defect detection — A review [J]. Image and Vision Computing, 2011, 29(7):442-458.
[2] i^, ^sra, SiMMItt^MH^&lffift&M [J]. 2019, 27(5):57-61.
[3] Zou Z.X., Chen K.Y., Shi Z.W., et al. Object detection in 20 years: A survey [J]. Proceedings of the IEEE, 2023, 111(3):257-276.
[4] Viola P., Jones M. Rapid object detection using a boosted cascade of simple features [C] // Proceedings of the 2001 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2001, 1:I-511-I-518.
[5] Dalal N., Triggs B. Histograms of oriented gradients for human detection [C] // Proceedings of the 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE,2005:886-893.
[6] Felzenszwalb P., Mcallester D., Ramanan D. A discriminatively trained, multiscale, deformable part model [C] // Proceedings of the 2008 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2008:1-8.
[7] Redmon J., Farhadi A. Y0L09000: better, faster, stronger [C] // Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2017:6517-6525.
[8] Redmon J., Farhadi A. Y0L0v3: an incremental improvement [EB/OL]. (2018-4-8).
[9] Bochkovskiy A., Wang C.Y., Liao H.Y.M. Y0L0v4: optimal speed and accuracy of object detection [EB/OL]. (2020-4-23)[2023-5-29].
[10] Nelson J., Solawetz J. Y0L0v5 is here: state-of-the-art object detection at 140 FPS[EB/0L]. (2020-6-10).
[11] Li C., Li L., Jiang H., et al. Y0L0v6: a single-stage object detection framework for industrial applications [EB/OL]. (2022-9-7).
[12] Wang C.Y., Bochkovskiy A., Liao H.Y.M. Y0L0v7: trainable bag-of-freebies sets new state-of-the-art for real-time object detectors [EB/OL]. (2022-7-6)
[13] Solawetz J., Francesco. What is Y0L0v8? The ultimate guide [EB/OL]. (2023-1-11).
[14] Zhang H., Zhang L., Li P., et al. Yarn-dyed fabric defect detection with Y0L0V2 based on deep convolution neural networks [C] // 2018 IEEE 7th Data Driven Control and Learning Systems Conference (DDCLS). Enshi: IEEE, 2018:170-174.
[15] ili. [D]. ttSH^f, 2021. 1.
[16] [D].
2021.
[17] S^ Y0L0v5-DCN l^tt^MH [J]. 2023, 51(03):8-14.
[18] «M, M«, Y0L0v7 [J/OL]. 2023:1-7.
References
[1] Henry Y., Grantham K.P., Nelson H.Y. Automated fabric defect detection — A review [J]. Image and Vision Computing, 2011, 29(7):442-458.
[2] Wang Mengtao, Li Yueyang, Du Shuai. Research progress of defect detection method based on machine vision [J]. Modern Textile Technology, 2019, 27(5):57-61.
[3] Zou Z.X., Chen K.Y., Shi Z.W., et al. Object detection in 20 years: A survey [J]. Proceedings of the IEEE, 2023, 111(3):257-276.
[4] Viola P., Jones M. Rapid object detection using a boosted cascade of simple features [C] // Proceedings of the 2001 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2001, 1:I-511-I-518.
[5] Dalal N., Triggs B. Histograms of oriented gradients for human detection [C] // Proceedings of the 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE,2005:886-893.
[6] Felzenszwalb P., Mcallester D., Ramanan D. A discriminatively trained, multiscale, deformable part model [C] // Proceedings of the 2008 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2008:1-8.
[7] Redmon J., Farhadi A. Y0L09000: better, faster, stronger [C] // Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2017:6517-6525.
[8] Redmon J., Farhadi A. Y0L0v3: an incremental improvement [EB/OL]. (2018-4-8).
[9] Bochkovskiy A., Wang C.Y., Liao H.Y.M. Y0L0v4: optimal speed and accuracy of object detection [EB/0L]. (2020-4-23)[2023-5-29].
[10] Nelson J., Solawetz J. Y0L0v5 is here: state-of-the-art object detection at 140 FPS[EB/0L]. (2020-6-10).
[11] Li C., Li L., Jiang H., et al. Y0L0v6: a single-stage object detection framework for industrial applications [EB/0L]. (2022-9-7).
[12] Wang C.Y., Bochkovskiy A., Liao H.Y.M. Y0L0v7: trainable bag-of-freebies sets new state-of-the-art for real-time object detectors [EB/0L]. (2022-7-6)
[13] Solawetz J., Francesco. What is Y0L0v8? The ultimate guide [EB/0L]. (2023-1-11).
[14] Zhang H., Zhang L., Li P., et al. Yarn-dyed fabric defect detection with Y0L0V2 based on deep convolution neural networks [C] // 2018 IEEE 7th Data Driven Control and Learning Systems Conference (DDCLS). Enshi: IEEE, 2018:170-174.
[15] Peng Yannan. Research on fabric blemish detection based on deep learning [D]. Jiangxi University of Technology, 2021. 1.
[16] Xu Henghui. Algorithm Research on High-Speed 0nline Detection of Multi-Texture Fabric Defects Based on Deep Learning [D]. Guilin University of Technology, 2021.
[17] Hu Yuejie,JIANG Gaoming. Fabric defect detection based on Y0L0v5-DCN [J]. Cotton Textile Technology, 2023, 51(03):8-14.
[18] Guo Dianpeng,KE Haisen,LI Xiaolu et al. Fabric defect detection algorithm based on improved Y0L0v7 [J]. Cotton Textile Technology: 1-7.
