基于B-YOLOv5的轻量化裂缝检测算法

胡肖肖; 雷斌; 蒋林; 段宛妮

doi:10.11973/wsjc202407010

基于B-YOLOv5的轻量化裂缝检测算法

胡肖肖^{1, 2,},
雷斌^{1, 2, ,},
蒋林^{1, 2},
段宛妮^{1, 2}

1.
武汉科技大学机械自动化学院,武汉 430081
2.
武汉科技大学机器人与智能系统研究院,武汉 430081

基金项目:

国家重点研发计划 2019YFB1310000

详细信息

作者简介:
胡肖肖（1998—）,男,硕士研究生,研究方向为计算机视觉,839172936@qq.com

通讯作者:
雷斌（1979—）,男,副教授,博士,研究方向为机器视觉、群体机器人,leibin@wust.edu.cn

中图分类号: TP391;TG115.28
计量
- 文章访问数: 30
- HTML全文浏览量: 12
- PDF下载量: 10
出版历程
- 收稿日期: 2023-12-18
- 刊出日期: 2024-07-09

Lightweight crack detection algorithm based on B-YOLOv5

HU Xiaoxiao^{1, 2,},
LEI Bin^{1, 2, ,},
JIANG Lin^{1, 2},
DUAN Wanni^{1, 2}

1.
College of Machinery and Automation, Wuhan University of Science and Technology, Wuhan 430081, China
2.
Institute of Robotics and Intelligent Systems, Wuhan University of Science and Technology, Wuhan 430081, China

摘要

摘要:
针对当前公路路面缺陷检测算法存在的特征提取不完善且难以部署到嵌入式设备上、细小裂纹及凹坑漏检等问题,以YOLOv5算法为基础,使用DepthSepConv模块代替原有的C3结构,把原有的CSPDarknet53主干网络改进成了更加轻量化的网络结构,结合BIFPN特征融合思想,将原来的PANet路径融合结构改进为一种更有效的带权重的B-PANet特征融合结构。试验结果表明,所改进的B-YOLOv5算法在相同的数据集和试验条件下,不仅精度提高了5.81%、检测速度提升两倍,还可降低细小裂纹和凹坑的漏检率,模型参数大小也仅仅是YOLOv5的八分之一。B-YOLOv5算法完全可以满足实时性的需要,且可更好地部署在Jetson Xavier NX嵌入式设备上。
- 裂缝检测 /
- BIFPN /
- YOLOv5 /
- 特征融合 /
- B-YOLOv5
Abstract:
In view of the current highway pavement defect detection algorithm feature extraction is imperfect and difficult to deploy on embedded equipment, missing detection of tiny cracks and pits, this paper used the Depth Sep Conv module instead of the original C3 structure, the original CSP Darknet 53 backbone network was improved into more lightweight network structure by combining with BIFPN feature fusion ideas, the original PANet path fusion structure was improved to be a more effective weight B-PANet feature fusion structure. The experimental results showed that the B-YOLOv5 algorithm improved in this paper can not only improve the accuracy of 5.81% and double the detection speed under the same data set and experimental conditions, but also improve the missed detection problem of fine cracks and pits, and the parameter size of the model was only one eighth of YOLOv5. The B-YOLOv5 algorithm can fully meet the needs of real-time performance and be better deployed on embedded devices.
- crack detection /
- BIFPN /
- YOLOv5 /
- feature fusion /
- B-YOLOv5

HTML全文

图 1 YOLOv5的网络结构

下载: 全尺寸图片幻灯片

图 2 C3网络结构

下载: 全尺寸图片幻灯片

图 3 改进后的主干网络结构

下载: 全尺寸图片幻灯片

图 4 FPN+PANet路径聚合结构和B-PANet结构

下载: 全尺寸图片幻灯片

图 5 两种算法的损失收敛对比图

下载: 全尺寸图片幻灯片

图 6 两种算法的P-R曲线对比

下载: 全尺寸图片幻灯片

图 7 两种算法不同环境条件下的检测对比试验

下载: 全尺寸图片幻灯片

Table 1 两种算法的网络参数比较

方法 YOLOv5 YOLOv5-B-PANet B-YOLOv5 YOLOv8

参数节点/M 7.1 8.0 5.5 3.0
浮点运算总次数/G 16.4 17.1 2.7 8.2
精度/% 91.25 98.71 96.56 95.4
推理速度/ms 26 26 10 6.0

下载: 导出CSV

Table 2 两种算法在数据集上的运行结果

方法 DepthSepConv PANet B-PANet m_AP50

A × √ × 68.20%
B × × √ 81.56%
C √ √ × 77.56%
D √ × √ 80.50%

下载: 导出CSV

参考文献(15)

[1]	陈克鸿,陈冠雄.我国公路养护市场化的问题与发展建议[J].中国公路,2019（13）：46-49.
[2]	周基,蔡强,田琼.70年中国公路路基路面病害研究现状与发展趋势——基于CNKI 1949—2019年文献的知识图谱分析[J].中外公路,2020,40（3）：60-66.
[3]	黎蔚,朱平哲.沥青路面裂缝图像检测算法研究[J].计算机工程与应用,2012,48（19）：163-166,219.
[4]	孙波成,邱延峻,梁世庆.基于小波的路面裂缝识别研究[J].重庆交通大学学报（自然科学版）,2010,29（1）：69-72.
[5]	封晓晨,李宁,顾玉宛,等.基于改进U-Net网络的细小裂纹检测[J].计算机应用与软件,2022,39（3）：193-200.
[6]	KUMAR P ,SHARMA A ,KOTA S R.Automatic multiclass instance segmentation of concrete damage using deep learning model[J].IEEE Access,2021,9：90330-90345.
[7]	罗晖,余俊英,涂所成.基于深度学习的公路路面病害检测算法[J].科学技术与工程,2022,22（13）：5299-5305.
[8]	BRAZ J,PETTRÉ J,RICHARD P2015 IEEE international conference on computer visionICCV 2015Santiago,Chile[s.n]2015BRAZ J ,PETTRÉ J ,RICHARD P.2015 IEEE international conference on computer vision[C]//ICCV 2015,Santiago,Chile：[s.n],2015.
[9]	WANG C Y,MARK LIAO H Y,WU Y H,et alCSPNet：a new backbone that can enhance learning capability of CNN2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops（CVPRW）Seattle,WA,USA IEEE202015711580WANG C Y ,MARK LIAO H Y ,WU Y H ,et al.CSPNet：a new backbone that can enhance learning capability of CNN[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops（CVPRW）.Seattle,WA,USA: IEEE,2020：1571-1580.
[10]	LIU S,QI L,QIN H F,et alPath aggregation network for instance segmentation2018 IEEE/CVF Conference on Computer Vision and Pattern RecognitionSalt Lake City,UT,USA IEEE201887598768LIU 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．Salt Lake City,UT,USA: IEEE,2018：8759-8768.
[11]	LIN T Y,DOLLÁR P,GIRSHICK R,et alFeature pyramid networks for object detection2017 IEEE Conference on Computer Vision and Pattern Recognition（CVPR）Honolulu,HI,USAIEEE2017936944LIN 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）.Honolulu,HI,USA: IEEE,2017：936-944.
[12]	ZHANG X Y,ZHOU X Y,LIN M X,et alShuffleNet：an extremely efficient convolutional neural network for mobile devices2018 IEEE/CVF Conference on Computer Vision and Pattern RecognitionSalt Lake City,UT,USA IEEE201868486856ZHANG X Y ,ZHOU X Y ,LIN M X ,et al.ShuffleNet：an extremely efficient convolutional neural network for mobile devices[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition.Salt Lake City,UT,USA: IEEE,2018：6848-6856.
[13]	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.
[14]	TAN M X,PANG R M,LE Q VEfficientDet：scalable and efficient object detection2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition（CVPR）Seattle,WA,USA IEEE20201077810787TAN M X ,PANG R M ,LE Q V.EfficientDet：scalable and efficient object detection[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition（CVPR）．Seattle,WA,USA: IEEE,2020：10778-10787.
[15]	LIN T Y,GOYAL P,GIRSHICK R,et alFocal loss for dense object detection2017 IEEE International Conference on Computer Vision（ICCV）Venice IEEE201729993007LIN T Y ,GOYAL P ,GIRSHICK R ,et al.Focal loss for dense object detection[C]//2017 IEEE International Conference on Computer Vision（ICCV）.Venice:  IEEE,2017：2999-3007.