基于机器视觉的浆纱毛羽量在线检测方法

doi:10.13475/j.fzxb.20250706701

摘要/Abstract

摘要：

为实现浆纱毛羽在线检测,提出了一种基于机器视觉的浆纱毛羽在线检测方法,包括在线图像采集装置和图像分割模型与毛羽评价指标。首先由工业相机、激光光源、电控丝杆等组成的在线图像采集装置,经自动对焦后连续采集上浆后片纱高清图像;随后将图像输入到所构建的图像分割深度学习模型得到毛羽分割二值图像,并采用阈值分割算法得到纱线主干二值图像;最后分别统计毛羽二值图像和纱线主干二值图像中的边缘像素总数,计算二者比值作为毛羽量指标输出。实验结果表明:所提方法对不同组分、线密度、颜色的纱线毛羽量指标与H指数相关性达94%,回归分析决定系数R²值为0.87,检测时效性超过10帧/s,具有良好的检测精度、实时性和品种普适性;同时,在不同上浆车速及照明亮度下测得结果偏差不超过5%,表明该方法能够适应上浆生产复杂条件并满足毛羽实时检测的需要。

关键词: 浆纱毛羽, 机器视觉, 在线检测, 图像分割, 毛羽评价指标

Abstract:

Objective Sizing is a key step in weaving preparation. It effectively controls yarn hairiness, improves weaving efficiency, and enhances fabric quality. However, hairiness detection during sizing still relies on offline methods. These methods fail to provide timely feedback on hairiness levels and hinder precise process control. To address this issue, we proposed a machine vision-based method for online detection of sized yarn hairiness. This method is expected to enable real-time hairiness monitoring during the sizing process.

Method An online image acquisition system was built using an industrial camera, a laser light source, and a motorized screw rod. After autofocus, high-resolution images of the sized yarn sheet were continuously captured. The acquired images were then processed by a deep-learning-based segmentation model to generate binary hairiness images. A thresholding algorithm was applied to extract binary yarn images. Edge pixels in both the hairiness and yarn stem images were counted separately. The hairiness ratio (H_r) was calculated as the ratio of edge pixels in the hairiness binary image to those in the yarn core binary image, serving as an indicator for evaluating the sized yarn hairiness level.

Results To verify the effectiveness of the proposed method under various influencing factors, an experimental device was set up and used to simulate the actual yarn running conditions in the sizing process. A variety of yarn samples with different linear densities, fiber types, colors, and construction type were selected to ensure the validity and universality of proposed method. The detection results (H_r) were compared with the H index measured by the Uster Tester 5. The detection results showed that the proposed method achieved the a correlation coefficient of 0.94 between H_r and the H index, outperforming the UNet model's 0.91 and the traditional Canny algorithm's 0.83. In terms of efficiency, both the proposed method and the UNet model, as deep learning-based approaches, were able to fully utilize the parallel computing capabilities GPU, and the number of images processed per second was significantly higher than the 2.87 frames per second of the Canny algorithm. Benefiting from lightweight design, the proposed method achieved a processing speed of over 10 frames per second, which is superior to the 7.31 frames per second of UNet, indicateing improved detection accuracy, real-time performance, and variety adaptability. To assess the robustness and operational stability of the proposed method under dynamic production conditions, a speed adaptability test was conducted. Hr values were measured at yarn running speeds of 70, 90, and 110 m/min and compared with baseline results at 50 m/min. Test results showed that H_r initially decreased with increasing speed but rose again at higher speeds. This trend is attributed to the edge diffusion effect, where blurred or diffused hairiness boundaries increase the number of detected edge pixels, raising the H_r value. Despite the influence of blur caused by motion, deviations in detection results across all tested speeds remained within 5%. In terms of environmental adaptability, the detection result of the proposed method at illumination levels above level 7 was only 5% compared to level 10 brightness. This proves that the method can adapt to the real-time detection of hairiness under complex conditions in sizing production.

Conclusion This study developed an online hairiness detection system on sizing machine, including an image acquisition device, a deep learning segmentation model, and a new index (H_r) to characterize the hairiness. The results showed that the H_r and the H index had a high level of correlation, which confirms the effectiveness of the proposed method. The experiments demonstrated that the proposed method had desirable effectiveness and real-time efficiency, as well as sufficient adaptability under varying speeds, illumination conditions, and yarn properties. During the sizing process, the method can record and plot the variation curve of hairiness, allowing real-time monitoring and data archiving. The hairiness detection results also reflect the influence of process parameters in real time and provides historical data to support process optimization.

Key words: sized yarn hairiness, machine vision, online detection, image segmentation, hairiness evaluation index

中图分类号:

TS103.7

黄昱翔, 潘欣明, 郭明瑞, 王静安, 高卫东. 基于机器视觉的浆纱毛羽量在线检测方法[J]. 纺织学报, 2026, 47(04): 136-144.

HUANG Yuxiang, PAN Xinming, GUO Mingrui, WANG Jing'an, GAO Weidong. Online detection of sized yarn hairiness based on machine vision[J]. Journal of Textile Research, 2026, 47(04): 136-144.

导出引用管理器 EndNote|Reference Manager|ProCite|BibTeX|RefWorks

链接本文: http://www.fzxb.org.cn/CN/10.13475/j.fzxb.20250706701

http://www.fzxb.org.cn/CN/Y2026/V47/I04/136

图/表 18

图1

图2

图3

图4

图5

图6

图7

图8

图9

表1

表2

表3

图10

表4

表5

表6

表7

图11

参考文献 25

[1]	秦少雄. 纱线毛羽检测仪器的研究及指标探讨[J]. 纺织导报, 2021(11): 58-61.
	QIN Shaoxiong. Research and index discussion of yarn hairiness tester[J]. China Textile Leader, 2021(11): 58-61.
[2]	HALEEM N, WANG X G. A comparative study on yarn hairiness results from manual test and two commercial hairiness metres[J]. The Journal of the Textile Institute, 2013, 104(5): 494-501. doi: 10.1080/00405000.2012.745235
[3]	马珂, 严凯, 张缓缓, 等. 基于贝叶斯阈值的纱线毛羽检测方法研究[J]. 棉纺织技术, 2021, 49(4): 11-15.
	MA Ke, YAN Kai, ZHANG Huanhuan, et al. Study on yarn hairiness measurement method based on Bayesian threshold[J]. Cotton Textile Technology, 2021, 49(4): 11-15.
[4]	李鹏飞, 严凯, 张缓缓, 等. 基于最大熵与密度聚类相融合的毛羽检测[J]. 纺织学报, 2019, 40(7): 158-162.
	LI Pengfei, YAN Kai, ZHANG Huanhuan, et al. Hairiness detection based on maximum entropy and density clustering[J]. Journal of Textile Research, 2019, 40(7): 158-162. doi: 10.1177/004051757004000209
[5]	CAO H Y, CHEN Z Z, HU H H, et al. Yarn hairiness measurement based on multi-camera system and perspective maximization model[J]. Journal of Electronic Imaging, 2024, 33(4): 043043.
[6]	JING J F, HUANG M Y, LI P F, et al. Automatic measurement of yarn hairiness based on the improved MRMRF segmentation algorithm[J]. The Journal of the Textile Institute, 2018, 109(6): 740-749. doi: 10.1080/00405000.2017.1368106
[7]	QIU Z J, WANG J G, GAO W D. A computer-vision-based nondestructive hairiness measurement system on yarn cone transportation line[J]. IEEE Transactions on Instrumentation and Measurement, 2024, 73: 3512513.
[8]	WANG J G, XU B G, LI Z J, et al. Depth recovery of hairy fibers for precise yarn hairiness measurement[J]. Applied Optics, 2018, 57(24): 7021-7029. doi: 10.1364/AO.57.007021
[9]	DENG Z M, YU L, WANG L J, et al. An algorithm for cross-fiber separation in yarn hairiness image processing[J]. The Visual Computer, 2024, 40(5): 3591-3599. doi: 10.1007/s00371-023-03053-z
[10]	YANG C H, XU Y, ZHANG J P, et al. Evenness and hairiness measurement of chenille yarn based on an improved pixel difference convolution for the pixel difference network[J]. Journal of Engineered Fibers and Fabrics, 2025, 20: 15589250251338256.
[11]	PEREIRA F, LOPES H, PINTO L, et al. A novel deep learning approach for yarn hairiness characterization using an improved YOLOv5 algorithm[J]. Applied Sciences, 2024, 15(1): 149. doi: 10.3390/app15010149
[12]	PEREIRA F, LOPES H, PINTO L, et al. Yarn quality analysis by using computer vision and deep learning techniques[J]. Textile Research Journal, 2026, 96(3/4): 240-265. doi: 10.1177/00405175251331205
[13]	DAI J C, REN J, LI S J. PHL-YOLO: a real-time lightweight yarn inspection method[J]. Journal of Real-Time Image Processing, 2024, 22(1): 20. doi: 10.1007/s11554-024-01595-6
[14]	PEREIRA F, MACEDO A, PINTO L, et al. Intelligent computer vision system for analysis and characterization of yarn quality[J]. Electronics, 2023, 12(1): 236. doi: 10.3390/electronics12010236
[15]	BAO J M, JING J F, XIE Y H. A defect detection system of glass tube yarn based on machine vision[J]. Journal of Industrial Textiles, 2023, 53: 15280837231152878.
[16]	TANG J L, YANG Z L, ZHANG X Y, et al. Intelligent detection method of yarn hairiness using CenterNet-CBAM[C]// BALAS V E, WU Q.Design studies and intelligence engineering. Amsterdam: IOS Press, 2022: 218-226.
[17]	CHOLLET F. Xception:deep learning with depthwise separable convolutions[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). New York: IEEE, 2017: 1800-1807.
[18]	AKRAM F, GARCIA M A, PUIG D. Active contours driven by difference of Gaussians[J]. Scientific Reports, 2017, 7: 14984. doi: 10.1038/s41598-017-14502-w pmid: 29101392
[19]	GUO M R, GAO W D, WANG J G. Online measurement of sizing yarn hairiness based on computer vision[J]. Fibers and Polymers, 2023, 24(4): 1539-1552. doi: 10.1007/s12221-023-00136-5
[20]	YANG P, SONG W, ZHAO X B, et al. An improved Otsu threshold segmentation algorithm[J]. International Journal of Computational Science and Engineering, 2020, 22(1): 146. doi: 10.1504/IJCSE.2020.107266
[21]	RONNEBERGER O, FISCHER P, BROX T. U-Net: convolutional networks for biomedical image segmentation[M]//Medical image computing and computer-assisted intervention-MICCAI 2015. Cham: Springer International Publishing, 2015: 234-241.
[22]	ROMERA E, ÁLVAREZ J M, BERGASA L M, et al. ERFNet: efficient residual factorized ConvNet for real-time semantic segmentation[J]. IEEE Transactions on Intelligent Transportation Systems, 2018, 19(1): 263-272. doi: 10.1109/TITS.2017.2750080
[23]	JIANG J T, WANG M M, TIAN H Z, et al. LV-UNet:a lightweight and vanilla model for medical image segmentation[C]//2024 IEEE International Conference on Bioinformatics and Biomedicine (BIBM). New York: IEEE, 2025: 4240-4246.
[24]	CANNY J. A computational approach to edge detection[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1986, PAMI-8(6): 679-698. doi: 10.1109/TPAMI.1986.4767851
[25]	XIE K, HOLDER S, GÖPFERT A, et al. An experimental study of motion blur in optical coordinate metrology for dynamic measurements of geometrical features[C]// Proceedings of Joint International IMEKO TC1+TC7+TC13 Symposium. Jena: IMEKO, 2011.

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed

纱线样本编号	纱线属性					H 指数
纱线样本编号	颜色	原料	原纱/ 浆纱	线密度/tex		H 指数
1-1	白色	棉	原纱	14.6	4.33
1-2	白色	棉	浆纱	14.6	2.52
1-3	白色	棉	浆纱	14.6	2.03
1-4	白色	棉	浆纱	14.6	1.89
1-5	白色	棉	浆纱	14.6	1.63
2-1	白色	棉	原纱	27.8	6.81
2-2	白色	棉	浆纱	27.8	2.97
3-1	白色	涤纶/棉(60/40)	原纱	13.0	2.90
3-2	白色	涤纶/棉(60/40)	浆纱	13.0	1.87
4-1	白色	粘胶	原纱	14.6	3.70
4-2	白色	粘胶	浆纱	14.6	2.17
5-1	白色	涤纶	原纱	14.6	6.82
5-2	白色	涤纶	浆纱	14.6	5.36
6-1	白色	腈纶	原纱	19.7	8.39
6-2	白色	腈纶	浆纱	19.7	5.80
7-1	白色	锦纶	原纱	14.6	5.19
7-2	白色	锦纶	浆纱	14.6	2.95
8-1	蓝色	棉	原纱	14.8×2	7.46
8-2	蓝色	棉	浆纱	14.8×2	3.27
9-1	绿色	棉	原纱	18.2×2	7.10
9-2	绿色	棉	浆纱	18.2×2	3.34
10-1	咖啡	棉	原纱	29.2×2	9.23
10-2	咖啡	棉	浆纱	29.2×2	4.23

k₁	δ₁	k₂	δ₂	TIoU
3	0.8	7	1.8	0.79
5	1.0	7	1.8	0.80
5	1.0	9	2.4	0.80
7	1.8	11	2.8	0.82
9	2.4	11	2.4	0.80
无高斯模糊				0.78

纱线样本编号	H 指数	H_r
纱线样本编号	H 指数	LFMHNet	UNet	ERFNet	LV_UNet	Canny
1-1	4.33	2.39	2.65	2.39	2.46	2.59
1-2	2.52	0.54	0.68	0.66	0.89	0.82
1-3	2.03	0.24	0.35	0.62	0.56	0.52
1-4	1.89	0.47	0.45	0.65	0.64	0.75
1-5	1.63	0.34	0.41	0.31	0.59	0.75
2-1	6.81	3.65	3.55	2.95	3.81	3.22
2-2	2.97	0.80	0.92	1.14	1.26	1.16
3-1	2.90	1.92	1.86	1.31	1.11	2.38
3-2	1.87	0.45	0.62	0.53	0.95	0.77
4-1	3.70	1.74	1.64	1.96	1.91	2.21
4-2	2.17	0.78	0.56	1.32	1.21	1.23
5-1	6.82	3.08	2.34	2.84	2.82	2.34
5-2	5.36	2.35	1.35	2.64	2.99	2.75
6-1	8.39	3.35	3.60	2.93	2.88	2.60
6-2	5.80	1.57	1.45	1.53	1.60	1.45
7-1	5.19	2.17	2.54	2.22	2.17	3.17
7-2	2.95	1.21	0.69	1.54	1.87	0.69
8-1	7.46	3.17	3.25	3.43	3.29	3.61
8-2	3.27	0.45	0.65	0.79	0.51	0.55
9-1	7.10	3.50	3.65	3.98	3.68	3.91
9-2	3.34	0.67	0.64	0.42	0.48	0.46
10-1	9.23	4.88	4.88	4.15	4.62	4.13
10-2	4.23	1.13	0.79	0.76	0.71	0.85
相关系数		0.94	0.91	0.90	0.89	0.83

图像处理方法	图像尺寸/像素	图像处理速度/(帧·s^-1)
LFMHNet	2 448×2 048	10.12
UNet	2 448×2 048	7.31
ERFNet	2 448×2 048	6.75
LV_UNet	2 448×2 048	5.32
Canny	2 448×2 048	2.87

车速/(m·min^-1)	H_r	平均绝对百分比误差/%
50	2.39	0
70	2.37	0.84
90	2.32	2.93
110	2.50	4.60