基于改进YOLOv5的螺纹钢表面缺陷检测

doi:10.11996/JG.j.2095-302X.2023030427

图学学报 ›› 2023, Vol. 44 ›› Issue (3): 427-437.DOI: 10.11996/JG.j.2095-302X.2023030427

• 图像处理与计算机视觉 • 上一篇下一篇

基于改进YOLOv5的螺纹钢表面缺陷检测

胡欣¹(), 周运强¹, 肖剑²(), 杨杰²

1.长安大学能源与电气工程学院，陕西西安 710064
2.长安大学电子与控制工程学院，陕西西安 710064

收稿日期:2022-11-02 接受日期:2023-01-02 出版日期:2023-06-30 发布日期:2023-07-03
通讯作者: 肖剑(1975-)，男，副教授，博士。主要研究方向为信号处理、人工智能应用、模式识别和计算机视觉等。E-mail：xiaojian@chd.edu.cn
作者简介:
胡欣(1975-)，女，副教授，博士。主要研究方向为能源管理、计算机视觉和机器学习等。E-mail：huxin@chd.edu.cn
基金资助:
陕西省重点研发计划项目(2021GY-054);陕西省重点研发计划项目(2023-YBGY-094);宁夏回族自治区重点研发计划项目(2022BEG03072)

Surface defect detection of threaded steel based on improved YOLOv5

HU Xin¹(), ZHOU Yun-qiang¹, XIAO Jian²(), YANG Jie²

1. School of Energy and Electrical Engineering, Chang′an University, Xi′an Shaanxi 710064, China
2. School of Electronic and Control Engineering, Chang′an University, Xi′an Shaanxi 710064, China

Received:2022-11-02 Accepted:2023-01-02 Online:2023-06-30 Published:2023-07-03
Contact: XIAO Jian (1975-), associate professor, Ph.D. His main research interests cover signal processing, artificial intelligence applications, pattern recognition and computer vision, etc. E-mail：xiaojian@chd.edu.cn
About author:
HU Xin (1975-), associate professor, Ph.D. Her main research interests cover energy management, computer vision and machine learning, etc. E-mail：huxin@chd.edu.cn
Supported by:
Key R&D Project of Shaanxi Province(2021GY-054);Key R&D Project of Shaanxi Province(2023-YBGY-094);Key R&D Project of Ningxia Hui Autonomous Region(2022BEG03072)

摘要/Abstract

摘要：

针对在工业场景下螺纹钢表面缺陷检测精度低、漏检和误检率高等问题，提出了一种改进YOLOv5的螺纹钢表面缺陷检测算法。改进YOLOv5算法中，融合多空间金字塔池化模块(M-SPP)，优化网络，通过增加网络的深度加强特征的提取，可以一定程度上提高检测精度；添加改进的空间和坐标注意力模块(SCA)，进一步区分空间领域不同像素之间的权重关系，更加关注感兴趣的区域，减小非必要的区域权重，提高模型对小目标缺陷的关注度；使用双采样过渡模块(TB)进行下采样，减少重要特征的丢失，获取更多特征信息；利用k-means++算法重聚类锚框，生成的预设锚框更适应缺陷的不同尺度大小，提高算法的检测精度。通过在螺纹钢表面缺陷数据集上的实验结果表明，改进后的YOLOv5算法对螺纹钢表面缺陷检测具有良好的检测性能，优于其他对比的算法。改进YOLOv5算法的AP₅₀达到97.6%，相对于YOLOv5算法提高了3.2%，其他各项指标均有涨点，在保持原检测速度基本不变的情况下，精准地检测螺纹钢表面缺陷。

关键词: YOLOv5, 缺陷检测, 多空间金字塔, 注意力机制, 双采样过渡, k-means++

Abstract:

An improved YOLOv5 algorithm for surface defect detection was proposed to solve the problems of low detection accuracy, high missed detection, and false detection rate in industrial scenarios. The improved YOLOv5 algorithm incorporated the multi-space pyramid pooling module (M-SPP) to optimize the network and the detection accuracy could be improved to a certain extent by increasing the depth of the network for better feature extraction. The improved spatial and coordinate attention module (SCA) was introduced to further distinguish the weight relationship between different pixels in the spatial domain, put more emphasis on the region of interest. This algorithm reduced the unnecessary regional weight and enhanced the model’s attention to small target defects. The double sampling transition module (TB) was utilized for downsampling to reduce the loss of important features and obtain more feature information. The k-means ++ algorithm was also employed to reunite the class anchor frame, and the generated preset anchor frame was more suitable for different sizes of defects, thereby improving the detection accuracy of the algorithm. The experimental results on the surface defect dataset of spiral steel showed that the improved YOLOv5 algorithm achieved good detection performance for the surface defect detection of spiral steel, superior to other compared algorithms. The improved YOLOv5 algorithm achieved an AP₅₀ of 97.6%, 3.2% higher than the YOLOv5 algorithm, and all other indexes showed an increase. While maintaining the original detection speed, the algorithm could accurately detect the surface defects of steel rebar.

Key words: YOLOv5, defect detection, multi-spatial pyramid pooling, attention mechanism, double sampling transition, k-means++

中图分类号:

TP391

胡欣, 周运强, 肖剑, 杨杰. 基于改进YOLOv5的螺纹钢表面缺陷检测[J]. 图学学报, 2023, 44(3): 427-437.

HU Xin, ZHOU Yun-qiang, XIAO Jian, YANG Jie. Surface defect detection of threaded steel based on improved YOLOv5[J]. Journal of Graphics, 2023, 44(3): 427-437.

图/表 15

图1 YOLOv5网络结构

Fig. 1 YOLOv5 network structure

图2 改进后YOLOv5网络结构

Fig. 2 Network structure of YOLOv5 after improvement

图3 M-SPP结构

Fig. 3 M-SPP structure

图4 SCA网络结构

Fig. 4 SCA structure

图5 TB结构

Fig. 5 TB structure

图6 3种尺度Anchor

Fig. 6 Anchor of 3 scales

表1 实验软硬件配置

Table 1 Experimental software and hardware onfiguration

名称	实验配置
操作系统	ubantu20.04
编程语言	Python 3.8
深度学习框架	PyTorch 1.8.0
CPU	Intel Core i7-9700K
GPU	NVIDIA RTX 3050 (6 G)
Cuda	Cuda 11.2
平台	Pycharm 2022.1

图7 部分数据集((a)划伤；(b)结疤；(c)锈迹；(d)划伤和锈)

Fig. 7 Partial data set (a) Scratch; (b) Scar; (c) Rust; (d) Scratch and rust)

表2 消融实验结果对比

Table 2 Comparison of ablation test results

k-means++	M-SPP	SCA	TB	AP (%)	AP₅₀ (%)	AP₇₅ (%)	AP_S (%)	AP_M (%)	AP_L (%)
-	-	-	-	64.7	94.4	68.6	39.7	62.9	68.6
√	-	-	-	65.8	95.7	69.9	39.9	63.3	68.7
-	√	-	-	64.9	94.6	68.5	40.1	63.7	69.9
-	-	√	-	66.2	96.2	69.4	42.5	63.8	69.4
-	-	-	√	65.6	94.7	68.5	39.7	63.3	68.6
√	√	√	√	67.4	97.6	70.8	42.8	65.3	70.0

表3 SCA使用不同池化结果对比

Table 3 Comparison of SCA pooling results

YOLOv5	Parameters (M)	AP_S (%)	AP_M (%)	AP_L (%)
+CA (GAP)	180.5	41.5	62.9	69.4
+CA (GMP)	180.5	41.9	63.1	69.3
+SCA (Ours)	180.5	42.5	63.8	69.4

表4 YOLOv5加入不同注意力

Table 4 YOLOv5 adds different attention

YOLOv5	Parameters (M)	AP_S (%)	AP_M (%)	AP_L (%)
+SE	180.2	39.7	63.1	69.1
+CBAM	180.7	40.4	64.0	69.2
+CA	180.4	40.9	62.8	68.8
+SCA (Ours)	180.5	42.5	63.8	69.4

表5 不同网络性能对比

Table 5 Performance comparison of different networks

Model	Backbone	AP (%)	AP₅₀ (%)	AP₇₅ (%)	AP_S (%)	AP_M (%)	AP_L (%)
Faster R-CNN^[7]	ResNet-50	57.1	87.7	52.2	33.9	51.3	58.2
YOLOv4^[14]	CSPDark-53	59.8	88.2	56.3	34.6	53.6	62.8
FCOS^[26]	ResNet-50	61.7	89.6	59.1	32.9	54.3	65.2
ATSS^[27]	ResNet-50	63.5	90.7	61.7	33.1	54.8	61.7
YOLOv5	Focus-CSPDarkNet	65.3	94.4	68.6	39.7	62.9	68.6
YOLOv7^[16]	ELAN	66.7	96.5	71.3	41.9	64.4	70.2
YOLOv5*(Ours)	Focus-CSPDarkNet-MSPP	67.4	97.6	70.8	42.8	65.3	70.0

图8 不同注意力热力图结果对比((a)锈迹；(b)划伤；(c)结疤)

Fig. 8 Comparison of results of different attention thermogram ((a) Rust; (b) Scratch; (c) Scar)

图9 可视化结果对比((a)原图；(b) YOLOv5网络检测结果；(c)改进YOLOv5网络检测结果)

Fig. 9 Comparison of visualization results ((a) Original figure; (b) YOLOv5 network detection results; (c) Improved YOLOv5 network detection results)

图10 缺陷误检实验对比((a)原图；(b) YOLOv5网络检测结果；(c)改进YOLOv5网络检测结果)

Fig. 10 Comparison of defect misdetection experiments ((a) Original figure; (b) YOLOv5 network detection results; (c) Improved YOLOv5 network detection results)

参考文献 29

[1]	唐莺. 基于脉冲漏磁检测机理的缺陷检测研究[EB/OL]. [2022-10-02]. https://xueshu.baidu.com/usercenter/paper/show?paperid=143568429b51eeba2a36b11d8e06f050&site=xueshu_se&hitarticle=1.
	TANG Y. Research on defect detection based on pulse magnetic flux leakage testing mechanism[EB/OL]. [2022-10-02]. https://xueshu.baidu.com/usercenter/paper/show? paperid=143568429b51eeba2a36b11d8e06f050&site=xueshu_se&hitarticle=1. (in Chinese)
[2]	武新军, 张卿, 沈功田. 脉冲涡流无损检测技术综述[J]. 仪器仪表学报, 2016, 37(8): 1698-1712.
	WU X J, ZHANG Q, SHEN G T. Review on advances in pulsed eddy current nondestructive testing technology[J]. Chinese Journal of Scientific Instrument, 2016, 37(8): 1698-1712. (in Chinese)
[3]	索会迎. 超声波无损检测技术应用研究[D]. 南京: 南京邮电大学, 2012.
	SUO H Y. The application and research of ultrasonic non-destructive testing technology[D]. Nanjing: Nanjing University of Posts and Telecommunications, 2012. (in Chinese)
[4]	HORAUD R, CHARRAS J P. Automatic inspection and orientation of external screws[EB/OL]. [2022-09-18]. https://www.researchgate.net/publication/292553139_AUTOMATIC_INSPECTION_AND_ORIENTATION_OF_EXTERNAL_SCREWS.
[5]	SOUKUP D, HUBER-MÖRK R. Convolutional neural networks for steel surface defect detection from photometric stereo images[M]//Advances in Visual Computing. Cham: Springer International Publishing, 2014: 668-677.
[6]	赵月, 张运楚, 孙绍涵, 等. 基于深度学习的螺纹钢表面缺陷检测[J]. 计算机系统应用, 2021, 30(7): 87-94.
	ZHAO Y, ZHANG Y C, SUN S H, et al. Defect detection method of rebar based on deep learning[J]. Computer Systems & Applications, 2021, 30(7): 87-94. (in Chinese)
[7]	REN S Q, HE K M, GIRSHICK R, et al. Faster R-CNN: towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(6): 1137-1149. DOI PMID
[8]	HE K M, ZHANG X Y, REN S Q, et al. Spatial pyramid pooling in deep convolutional networks for visual recognition[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015, 37(9): 1904-1916. DOI PMID
[9]	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. New York: IEEE Press, 2021: 13708-13717.
[10]	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. New York: IEEE Press, 2018: 8759-8768.
[11]	REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: unified, real-time object detection[C]// 2016 IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2016: 779-788.
[12]	REDMON J, FARHADI A. YOLO9000: better, faster, stronger[C]// 2017 IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2017: 6517-6525.
[13]	REDMON J, FARHADI A. YOLOv3: an incremental improvement[EB/OL]. [2022-10-02]. https://arxiv.org/abs/1804.02767.
[14]	BOCHKOVSKIY A, WANG C Y, LIAO H Y M. YOLOv4: optimal speed and accuracy of object detection[EB/OL]. [2022-10-02]. https://arxiv.org/abs/2004.10934.
[15]	唐靓, 余明慧, 武明虎, 等. 基于改进YOLOv5的绝缘子缺陷检测算法[J]. 华中师范大学学报: 自然科学版, 2022, 56(5): 771-780.
	TANG J, YU M H, WU M H, et al. Insulator defect detection algorithm based on improved YOLOv5[J]. Journal of Central China Normal University: Natural Sciences, 2022, 56(5): 771-780. (in Chinese)
[16]	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[EB/OL]. [2022-10-02]. https://arxiv.org/abs/2207.02696.
[17]	胡欣, 马丽军. 基于YOLOv5的多分支注意力SAR图像舰船检测[J]. 电子测量与仪器学报, 2022, 36(8): 141-149.
	HU X, MA L J. Multi-branch attention SAR image ship detection based on YOLOv5[J]. Journal of Electronic Measurement and Instrumentation, 2022, 36(8): 141-149. (in Chinese)
[18]	HU J, SHEN L, ALBANIE S, et al. Squeeze-and-excitation networks[C]// IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2018: 7132-7141.
[19]	WOO S, PARK J, LEE J Y, et al. CBAM: convolutional block attention module[EB/OL]. [2022-10-02]. https://arxiv.org/abs/1807.06514.
[20]	HUANG G, LIU Z, VAN DER MAATEN L, et al. Densely connected convolutional networks[C]// 2017 IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2017: 2261-2269.
[21]	LIKAS A, VLASSIS N, J VERBEEK J. The global k-means clustering algorithm[J]. Pattern Recognition, 2003, 36(2): 451-461. DOI URL
[22]	ARTHUR D, VASSILVITSKII S. K-Means++: the advantages of careful seeding[EB/OL] [2022-10-02]. https://xueshu.baidu.com/usercenter/paper/show?paperid=5d168abdad2ea2ca220841fdd783f488.
[23]	东辉, 陈鑫凯, 孙浩, 等. 基于改进YOLOv4和图像处理的蔬菜田杂草检测[J]. 图学学报, 2022, 43(4): 559-569.
	DONG H, CHEN X K, SUN H, et al. Weed detection in vegetable field based on improved YOLOv4 and image processing[J]. Journal of Graphics, 2022, 43(4): 559-569. (in Chinese)
[24]	张伟康, 孙浩, 陈鑫凯, 等. 基于改进YOLOv5的智能除草机器人蔬菜苗田杂草检测研究[J]. 图学学报, 2023, 44(2): 346-356.
	ZHANG W K, SUN H, CHEN X K, et al. Research on weed detection in vegetable seedling fields based on the improved YOLOv5 intelligent weeding robot[J]. Journal of Graphics, 2023, 44(2): 346-356. (in Chinese)
[25]	ZHOU B L, KHOSLA A, LAPEDRIZA A, et al. Learning deep features for discriminative localization[C]// 2016 IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2016: 2921-2929.
[26]	TIAN Z, SHEN C H, CHEN H, et al. FCOS: fully convolutional one-stage object detection[C]// 2019 IEEE/CVF International Conference on Computer Vision. New York: IEEE Press, 2020: 9626-9635.
[27]	ZHANG S F, CHI C, YAO Y Q, et al. Bridging the gap between anchor-based and anchor-free detection via adaptive training sample selection[C]// 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2020: 9756-9765.
[28]	蒋镕圻, 彭月平, 谢文宣, 等. 嵌入scSE模块的改进YOLOv4小目标检测算法[J]. 图学学报, 2021, 42(4): 546-555.
	JIANG R Q, PENG Y P, XIE W X, et al. Improved YOLOv4 small target detection algorithm with embedded scSE module[J]. Journal of Graphics, 2021, 42(4): 546-555. (in Chinese)
[29]	胡俊, 顾晶晶, 王秋红. 基于遥感图像的多模态小目标检测[J]. 图学学报, 2022, 43(2): 197-204.
	HU J, GU J J, WANG Q H. Multimodal small target detection based on remote sensing image[J]. Journal of Graphics, 2022, 43(2): 197-204. (in Chinese)

基于改进YOLOv5的螺纹钢表面缺陷检测

Surface defect detection of threaded steel based on improved YOLOv5

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 15

参考文献 29

相关文章 15

编辑推荐

Metrics

本文评价

[1]	李利霞, 王鑫, 王军, 张又元 . 基于特征融合与注意力机制的无人机图像小目标检测算法 [J]. 图学学报, 2023, 44(4): 658-666.
[2]	郝帅, 赵新生, 马旭, 张旭, 何田, 侯李祥. 基于 TR-YOLOv5 的输电线路多类缺陷目标检测方法 [J]. 图学学报, 2023, 44(4): 667-676.
[3]	曹义亲 , 周一纬 , 徐露 . 基于 E-YOLOX 的实时金属表面缺陷检测算法 [J]. 图学学报, 2023, 44(4): 677-690.
[4]	李鑫, 普园媛, 赵征鹏, 徐丹, 钱文华 . 内容语义和风格特征匹配一致的艺术风格迁移 [J]. 图学学报, 2023, 44(4): 699-709.
[5]	余伟群, 刘佳涛, 张亚萍. 融合注意力的拉普拉斯金字塔单目深度估计 [J]. 图学学报, 2023, 44(4): 728-738.
[6]	李刚, 张运涛, 汪文凯, 张东阳. 采用DETR与先验知识融合的输电线路螺栓缺陷检测方法[J]. 图学学报, 2023, 44(3): 438-447.
[7]	毛爱坤, 刘昕明, 陈文壮, 宋绍楼. 改进YOLOv5算法的变电站仪表目标检测方法[J]. 图学学报, 2023, 44(3): 448-455.
[8]	郝鹏飞, 刘立群, 顾任远. YOLO-RD-Apple果园异源图像遮挡果实检测模型[J]. 图学学报, 2023, 44(3): 456-464.
[9]	罗文宇, 傅明月. 基于YoloX-ECA模型的非法野泳野钓现场监测技术[J]. 图学学报, 2023, 44(3): 465-472.
[10]	李雨, 闫甜甜, 周东生, 魏小鹏. 基于注意力机制与深度多尺度特征融合的自然场景文本检测[J]. 图学学报, 2023, 44(3): 473-481.
[11]	吴文欢, 张淏坤. 融合空间十字注意力与通道注意力的语义分割网络[J]. 图学学报, 2023, 44(3): 531-539.
[12]	谢国波, 贺笛轩, 何宇钦, 林志毅. 基于 P-CenterNet 的光学遥感图像烟囱检测[J]. 图学学报, 2023, 44(2): 233-249.
[13]	熊举举 , 徐杨 , 范润泽 , 孙少聪 . 基于轻量化视觉 Transformer 的花卉识别 [J]. 图学学报, 2023, 44(2): 271-279.
[14]	陈刚 , 张培基 , 龚冬冬 , 于俊清 . 火电厂监控视频安全服检测方法研究[J]. 图学学报, 2023, 44(2): 291-297.
[15]	成浪, 敬超. 基于改进 YOLOv7 的 X 线图像旋转目标检测[J]. 图学学报, 2023, 44(2): 324-334.