基于改进的YOLOv5金刚石线表面质量检测

doi:10.16088/j.issn.1001-6600.2022112106

摘要/Abstract

摘要： 金刚石线上固着的金刚石颗粒数量、位置分布和分布密度等是衡量金刚石线表面质量的重要参数,也是衡量金刚石线切割能力的重要指标。针对金刚石颗粒微小、密集、粘连等现象导致其特征难以提取、识别准确率低等问题,本文采用深度学习技术,提出一种基于改进的YOLOv5金刚石线表面质量检测方法。首先在图像处理阶段,使用阈值分割技术初步划分大、小颗粒;其次在主干网络部分,添加CA(coordinate attention)注意力模块,获取更高质量的粘连颗粒中单粒边界特征;再次设计C2(CA+CBL)模块,用特征融合保留不同层之间的语义信息,进而提升密集小目标的检测精度;最后用一个卷积代替CSP2_X结构,降低计算量,同时将不同尺度的输出特征图感受野缩小,避免颗粒特征虚化,从而影响颗粒的检测精度。实验表明,改进的网络模型,能够有效识别形状不一、大小不一、粘连和密集的金刚石颗粒图像,微粒平均精度(AP)达到83.80%,大颗粒平均精度达到90.70%,平均精度均值(mAP)达到87.20%。

关键词: 金刚石线, YOLOv5, 注意力机制, 小目标检测, 阈值分割

Abstract: The number, position distribution and distribution density of the diamond particles fixed on the diamond wire are important parameters to measure the surface quality of the diamond wire, and also important indicators to measure the cutting ability of the diamond wire. Aiming at the problems of small, dense and adhesive diamond particles, such as difficult to extract their features and low accuracy, a method of surface quality detection based on improved YOLOv5 diamond wire is proposed by using deep learning technology. First, in the image processing stage, the threshold segmentation technology is used to preliminarily divide large and small particles; Secondly, in the backbone network part, the CA (coordinated attention) attention module is added to obtain high-quality single-particle boundary features in the adhesive particles; The C2 (CA+CBL) module is designed again to preserve the semantic information between different layers by feature fusion, thus improving the detection accuracy of dense small objects; Finally, replace CSP2_X with a convolution structure, reduce the calculation amount, and reduce the receptive field of the output characteristic map of different scales to avoid the virtualization of particle characteristics, thus affecting the particle detection accuracy. Experiments show that the improved network model can effectively identify the images of diamond particles with different shapes, sizes, adhesion and density, the average accuracy (AP) of the particles, and the large particles are 83.80%, and 90.70%, respectively, and the mean average precision (mAP) is 87.20%.

Key words: diamond wire, YOLOv5, attention mechanism, small object detection, threshold segmentation

中图分类号: TP391.41

黄叶祺, 王明伟, 闫瑞, 雷涛. 基于改进的YOLOv5金刚石线表面质量检测[J]. 广西师范大学学报（自然科学版）, 2023, 41(4): 123-134.

HUANG Yeqi, WANG Mingwei, YAN Rui, LEI Tao. Surface Quality Detection of Diamond Wire Based on Improved YOLOv5[J]. Journal of Guangxi Normal University(Natural Science Edition), 2023, 41(4): 123-134.

参考文献

[1] 李军. 电镀金刚石线对硅晶体切割机理的研究[D]. 青岛: 青岛科技大学, 2020. DOI: 10.27264/d.cnki.gqdhc.2020.000648.
[2] LIU T Y, GE P Q, BI W B, et al. A new method of determining the slicing parameters for fixed diamond wire saw[J]. Materials Science in Semiconductor Processing, 2020, 120: 105252. DOI: 10.1016/j.mssp.2020.105252.
[3] CHEN C Y, SUN M, CHEN X Q, et al. Recent advances of silicon wafer cutting technology for photovoltaic industry[J].Metallurgical Research and Technology, 2021, 118(6): 616. DOI: 10.1051/metal/2021091.
[4] YIN Y K, GAO Y F, YANG C F. Sawing characteristics of diamond wire cutting sapphire crystal based on tool life cycle[J]. Ceramics International, 2021, 47(19): 26627-26634. DOI: 10.1016/j.ceramint.2021.06.070.
[5] 王飞阳. 基于机器视觉的金刚线在线质检技术[D]. 哈尔滨: 哈尔滨工业大学, 2014. DOI: 10.7666/d.D593040.
[6] 刘明宇, 佃松宜. 基于机器视觉的金刚线表面质量检测[J]. 四川大学学报(自然科学版), 2020, 57(5): 920-926. DOI: 10.3969/j.issn.0490-6756.2020.05.015.
[7] 张文晔. 基于机器视觉的金刚砂线颗粒检测技术的研究与应用[D]. 常州: 江苏理工学院, 2018: 22-31.
[8] 杨建新, 兰小平, 王波, 等. 基于深度学习的黄色工业金刚石检测方法[J]. 金刚石与磨料磨具工程, 2020, 40(6): 13-19. DOI: 10.13394/j.cnki.jgszz.2020.6.0003.
[9] CUNHA A, FERREIRA R, TRINDADE B et al. Production of a laser textured 316L stainless steel reinforced with CuCoBe + diamond composites by hot pressing: Influence of diamond particle size on the hardness and tribological behaviour[J]. Tribology International, 2020, 146: 106056. DOI: 10.1016/j.triboint.2019.106056.
[10] 周洪宇, 冉珉瑞, 李亚强, 等. 颗粒尺寸对金刚石/Al封装基板热物性的影响[J]. 金属学报, 2021, 57(7): 937-947. DOI: 10.11900/0412.1961.2020.00393.
[11] 全国磨料磨具标准化技术委员会. 超硬磨料制品电镀金刚石线: JB/T 12543—2015[S]. 北京: 机械工业出版社, 2015.
[12] GAO Y F, WEI W, WANG K, et al. Online quality inspection technology for electroplated diamond wire based on machine vision[C]// 2014 IEEE International Conference on Robotics and Biomimetics (ROBIO 2014). Piscataway, NJ: IEEE, 2014: 2455-2459. DOI: 10.1109/ROBIO.2014.7090708.
[13] 陕西科技大学. 一种基于机器视觉的金刚石线表面颗粒检测设备: CN202123279302.1[P]. 2022-06-07.
[14] 杨高科. 图像处理、分析与机器视觉:基于LabVIEW[M]. 北京:清华大学出版社,2018: 16-32.
[15] 孙俊, 宋佳, 武小红, 等. 基于改进Otsu算法的生菜叶片图像分割方法[J]. 江苏大学学报(自然科学版), 2018, 39(2): 179-184. DOI: 10.3969/j.issn.1671-7775.2018.02.010.
[16] 郝建军, 邴振凯, 杨淑华, 等. 采用改进YOLOv3算法检测青皮核桃[J].农业工程学报, 2022, 38(14): 183-190. DOI: 10.11975/j.issn.1002-6819.2022.14.021.
[17] FAN Y X, CHEN Y Y, CHEN X, et al. Estimating the aquatic-plant area on a pond surface using a hue-saturation-component combination and an improved Otsu method[J]. Computers and Electronics in Agriculture, 2021, 188: 106372. DOI: 10.1016/j.compag.2021.106372.
[18] ALMOTIRI J, ELLEITHY K, ELLEITHY A. A multi-anatomical retinal structure segmentation system for automatic eye screening using morphological adaptive fuzzy thresholding[J]. IEEE Journal of Translational Engineering in Health and Medicine, 2018, 6: 3800123. DOI: 10.1109/JTEHM.2018.2835315.
[19] 郭磊, 王邱龙, 薛伟, 等. 基于改进YOLOv5的小目标检测算法[J]. 电子科技大学学报, 2022, 51(2): 251-258. DOI: 10.12178/1001-0548.2021235.
[20] ZHU L L, GENG X, LI Z, et al. Improving YOLOv5 with attention mechanism for detecting boulders from planetary images[J]. Remote Sens, 2021, 13(18): 3776. DOI: 10.3390/rs13183776.
[21] 彭浩. 基于YOLOv5的无人机巡检图像绝缘子检测技术的研究[D]. 徐州: 中国矿业大学, 2021. DOI: 10.27623/d.cnki.gzkyu.2021.003019.
[22] LIN T Y, DOLLÁR P, GIRSHICK R, et al. Feature pyramid networks for object detection[C]// 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Los Alamitos, CA: IEEE Computer Society, 2017: 936-944. DOI: 10.1109/CVPR.2017.106.
[23] LIU S, QI L, QIN H F, et al. Path aggregation network for instance segmentation[C]// 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Los Alamitos, CA: IEEE Computer Society, 2018: 8759-8768. DOI: 10.1109/ CVPR.2018.00913.
[24] LIU Z, LIN Y T, CAO Y, et al. Swin transformer: hierarchical vision transformer using shifted windows[EB/OL]. (2021-08-17)[2022-11-21]. https://arxiv.org/abs/2103.14030. DOI: 10.48550/arXiv.2103.14030.
[25] XU X K, FENG Z J, CAO C Q, et al. An improved swin transformer-based model for remote sensing object detection and instance segmentation[J]. Remote Sensing, 2021, 13(23): 4779. DOI: 10.3390/rs13234779.
[26] VASWANI A, SHAZEER N, PARMAR N, et al. Attention is all you need[C]// Advances in Neural Information Processing Systems 30 (NIPS 2017). Red Hook, NY: Curran Associates Inc., 2017: 6000-6010.
[27] HOU Q B, ZHOU D Q, FENG J S. Coordinate attention for efficient mobile network design[C]// 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Los Alamitos, CA: IEEE Computer Society, 2021: 13708-13717. DOI: 10.1109/CVPR46437.2021.01350.
[28] 侯志强, 刘晓义, 余旺盛, 等. 使用GIoU改进非极大值抑制的目标检测算法[J]. 电子学报, 2021, 49(4): 696-705. DOI: 10.12263/DZXB.20200132.
[29] 郑恩壮, 钟宝江. 各向异性的多尺度边缘检测算法[J]. 激光与光电子学进展, 2022, 59(4): 0142002. DOI: 10.3788/LOP202259.0412002.
[30] KISANTA M, WOJNA Z, MURAWSKI J, et al. Augmentation for small object detection[EB/OL]. (2019-02-19)[2022-11-21]. https://arxiv.org/abs/1902.07296. DOI: 10.48550/arXiv.1902.07296.
[31] LIANG D K, CHEN X W, XU W, et al. TransCrowd: weakly-supervised crowd counting with transformers[J]. Science China (Information Sciences), 2022, 65(6): 160104. DOI: 10.1007/s11432-021-3445-y.
[32] SELVARAJU R R, COGSWELL M, DAS A, et al. Grad-CAM: visual explanations from deep networks via gradient-based localization[J]. International Journal of Computer Vision, 2020, 128(2): 336-359. DOI: 10.1007/ s11263-019-01228-7.
[33] 李健, 刘孔宇, 任宪盛, 等. 基于自适应阈值的Canny算法在MRI边缘检测中的应用[J]. 吉林大学学报(工学版), 2021, 51(2): 712-719. DOI: 10.13229/j.cnki.jdxbgxb20200839.
[34] 黄露. 基于机器视觉的编织袋表面缺陷检测系统开发[D]. 武汉: 湖北工业大学, 2021. DOI: 10.27131/d.cnki.ghugc. 2021.000494.

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed