基于改进YOLOv5的路面坑洼检测方法

doi:10.16180/j.cnki.issn1007-7820.2024.07.007

摘要/Abstract

摘要：

坑洼是一种常见的路面病害,会降低行车安全,准确快速地检测路面坑洼较为重要。针对现有坑洼检测方法在小目标和密集目标的场景下检测精度不高的问题,文中提出了一种改进YOLOv5(You Only Look Once version 5)模型。在YOLOv5的主干网络中引入CBAM(Convolutional Block Attention Module)来提高模型对关键特征的注意能力,将YOLOv5的损失函数改为EIoU(Efficient Intersection over Union)来提高模型对目标的检测精度。实验结果表明,所提模型能够在小目标和密集目标的场景下快速准确地检测路面坑洼,在开源数据集Annotated Potholes Image Dataset中的mAP(mean Average Precision)达到了82%,较较于YOLOv5和其他主流方法也有所提高。

关键词: 路面坑洼, 深度学习, YOLOv5, 注意力机制, CBAM注意力, 小目标检测, 密集目标检测, 损失函数

Abstract:

Pothole is a common road disease, it reduce driving safety, accurate and rapid detection of potholes is more important.In viewof the problem that the detection accuracy of existing pothole detection methods is not high in the scenario of small targets and dense targets, an improved YOLOv5(You Only Look Once version 5) model is proposed in this study.TheCBAM(Convolutional Block Attention Module) is introduced into YOLOv5's backbone network to improve the model's ability to pay attention to key features. The loss function of YOLOv5 is changed to EIoU(Efficient Intersection over Union) to improve the detection accuracy of the model.The experimental results show that the proposed model can detect Potholes quickly and accurately in the scenarios of small targets and dense targets, and the mAP(mean Average Precision) in the open source Annotated Potholes Image Dataset reaches 82%. Compared with YOLOv5 and other mainstream methods, it is also improved.

Key words: pavement potholes, deep learning, YOLOv5, attention mechanism, CBAM attention, small target detection, dense target detection, loss function

中图分类号:

TP391

何幸, 黄永明, 朱勇. 基于改进YOLOv5的路面坑洼检测方法[J]. 电子科技, 2024, 37(7): 53-59.

HE Xing, HUANG Yongming, ZHU Yong. Pavement Pothole Detection Method Based on Improved YOLOv5[J]. Electronic Science and Technology, 2024, 37(7): 53-59.

图/表 13

图1

图2

图3

图4

图5

表1

图6

表2

表3

图7

表4

图8

表5

参考文献 22

[1]	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.
	Luo Hui, Yu Junying, Tu Suocheng. Pavement defect detection algorithm based on deep learning[J]. Science Technology andEngineering, 2022, 22(13):5299-5305.
[2]	李良福, 孙瑞赟. 复杂背景下基于图像处理的桥梁裂缝检测算法[J]. 激光与光电子学进展, 2019, 56(6):112-122.
	Li Liangfu, Sun Ruiyun. Bridge crack detection algorithm based on image processing under complex background[J]. Laser and Optoelectronics Progress, 2019, 56(6):112-122.
[3]	王玲敏, 段军, 辛立伟. 引入注意力机制的YOLOv5安全帽佩戴检测方法[J]. 计算机工程与应用, 2022, 58(9):303-312. doi: 10.3778/j.issn.1002-8331.2112-0242
	Wang Lingmin, Duan Jun, Xin Liwei. YOLOv5 helmet wear detection method with introduction of attention mechanism[J]. Computer Engineering and Applications, 2022, 58(9):303-312. doi: 10.3778/j.issn.1002-8331.2112-0242
[4]	程长文, 陈玮, 陈劲宏, 等. 改进YOLO的口罩佩戴实时检测方法[J]. 电子科技, 2023, 36(2):73-80.
	Cheng Changwen, Chen Wei, Chen Jinhong, et al. YOLO-improve detection method of real-time mask wearing[J]. Electronic Science and Technology, 2023, 36(2):73-80.
[5]	Suong L K, Jangwoo K. Detection of potholes using a deep convolutional neural network[J]. Journal of Universal Computer Science, 2018, 24(9):1244-1257.
[6]	罗晖, 贾晨, 李健. 基于改进YOLOv4的公路路面病害检测算法[J]. 激光与光电子学进展, 2021, 58(14):336-344.
	Luo Hui, Jia Chen, Li Jian. Road surface disease detection algorithm based on improved YOLOv4[J]. Laser and Optoelectronics Progress, 2021, 58(14):336-344.
[7]	Wang W, Wu B, Yang S, et al. Road damage detection and classification with faster R-CNN[C]. Seattle: IEEE International Conference on Big Data, 2018:5220-5223.
[8]	张宁. 基于Faster R-CNN的公路路面病害检测算法的研究[D]. 南昌: 华东交通大学, 2019:59-63.
	Zhang Ning. Study on detection algorithm for road surface disease based on Faster R-CNN[D]. Nanchang: East China Jiaotong University, 2019:59-63.
[9]	黄璐, 毛晓艳, 杜航, 等. 基于深度学习网络的星表非结构化岩石目标辨识方法研究[J]. 空间控制技术与应用, 2021, 47(6):27-33.
	Huang Lu, Mao Xiaoyan, Du Hang, et al. Onstar catalog unstructured rock target identification method based on deep learning network[J]. Aerospace Control and Application, 2021, 47(6):27-33.
[10]	Zhu X, Lyu S, Wang X, et al. TPH-YOLOv5:Improved YOLOv5 based on transformer prediction head for object detection on drone-captured scenarios[C]. Montreal: Proceedings of the IEEE/CVF International Conference on Computer Vision, 2021:2778-2788.
[11]	Nepal U, Eslamiat H. Comparing YOLOv3,YOLOv4 and YOLOv5 for autonomous landing spot detection in faulty UAVs[J]. Sensors, 2022, 22(2):464-478.
[12]	许德刚, 王露, 李凡. 深度学习的典型目标检测算法研究综述[J]. 计算机工程与应用, 2021, 57(8):10-25. doi: 10.3778/j.issn.1002-8331.2012-0449
	Xu Degang, Wang Lu, Li Fan. Review of typical object detection algorithms for deep learning[J]. Computer Engineering and Applications, 2021, 57(8):10-25. doi: 10.3778/j.issn.1002-8331.2012-0449
[13]	Jia W, Xu S, Liang Z, et al. Real-time automatic helmet detection of motorcyclists in urban traffic using improved YOLOv5 detector[J]. IET Image Processing, 2021, 15(14):3623-3637.
[14]	Li Z Z. Road aerial object detection based on improved YOLOv5[C]. Beihai: International Conference on Computer, Big Data and Artificial Intelligence, 2022: 1-5.
[15]	冯凯, 张书雅, 李锦暄, 等. 基于卷积神经网络的返回舱识别[J]. 现代信息科技, 2021, 5(10):20-26.
	Feng Kai, Zhang Shuya, Li Jinxuan, et al. Identification ofreturn cabin based on convolutional neural network[J]. Modern Information Technology, 2021, 5(10):20-26.
[16]	张伟, 刘娜, 江洋, 等. 基于YOLO神经网络的垃圾检测与分类[J]. 电子科技, 2022, 35(10):45-50.
	Zhang Wei, Liu Na, Jiang Yang, et al. Garbage detection and classification based on YOLO neural network[J]. Electronic Science and Technology, 2022, 35(10):45-50.
[17]	Woo S, Park J, Lee J Y, et al. Cbam:Convolutional block attention module[C]. Munich: Proceedings of the European Conference on Computer Vision, 2018:3-19.
[18]	Zhang Y F, Ren W, Zhang Z, et al. Focal and efficient IOU loss for accurate bounding box regression[J]. Neurocomputing, 2022, 50(6):146-157.
[19]	马志强. 基于改进YOLOv5的海珍品目标检测算法[J]. 现代信息科技, 2021, 5(18):80-85.
	Ma Zhiqiang. Sea treasure target detection algorithm based on improved YOLOv5[J]. Modern Information Technology, 2021, 5(18):80-85.
[20]	Park S S, Tran V T, Lee D E. Application of various yolo models for computer vision-based real-time pothole detection[J]. Applied Sciences, 2021, 11(23):11217-11229.
[21]	Bučko B, Lieskovská E, Zábovská K, et al. Computer vision based pothole detection under challenging conditions[J]. Sensors, 2022, 22(22):8871-8878.
[22]	Heo D H, Choi J Y, Kim S B, et al. Image-based pothole detection using multiscale feature network and risk assessment[J]. Electronics, 2023, 12(4):810-826.

序号	输入来源	模块大小	模块名称	模块参数
0	-1	3 520	Focus	[3,32,3]
1	-1	19 170	CBAM	[32,64,3,2]
2	-1	18 816	C3	[64,64,1]
3	-1	76 130	CBAM	[64,128,3,2]
4	-1	156 928	C3	[128,128,3]
5	-1	303 714	CBAM	[128,256,3,2]
6	-1	625 152	C3	[256,256,3]
7	-1	1 180 672	Conv	[256,512,3,2]
8	-1	656 896	SPP	[512,512,[5,9,13]]
9	-1	1 182 720	C3	[512,512,1,False]
10	-1	131 584	Conv	[512,256,1,1]
11	-1	0	Upsample	[None,2,’nearest’]
12	[-1,6]	0	Concat	[1]
13	-1	361 984	C3	[512,256,1,False]
14	-1	33 024	Conv	[256,128,1,1]
15	-1	0	Upsample	[None,2,’nearest’]
16	[-1,4]	0	Concat	[1]
17	-1	90 880	C3	[256,128,1,False]
18	-1	147 712	Conv	[128,128,3,2]
19	[-1,14]	0	Concat	[1]
20	-1	296 448	C3	[256,256,1,False]
21	-1	590 336	Conv	[256,256,3,2]
22	[-1,10]	0	Concat	[1]
23	-1	1 182 720	C3	[512,512,1,False]

实验环境	配置
操作系统	Linux Ubuntu 18.04
处理器	Intel(R) Xeon(R) CPU E5-2678 v3 @2.50 GHz
内存	8 GB
显卡	NVIDIA Tesla K80(12 GB)
显卡加速	cuda11.1 cuDNN 8.0.5
开发环境	Python 3.7
深度学习框架	Pytorch 1.8.1

参数名	参数值
LEANING_RATE(学习率)	0.010 0
MOMENTUM(冲量)	0.937 0
WEIGHT_DECAY(权重衰减)	0.000 5
BATCH_SIZE(批大小)	16
EPOCH(迭代次数)	100

对比模型	mAP/%	帧速率/frame·s^-1
Sparse R-CNN	72.6	-
YOLOv3	77.1	28.57
YOLOv3-SPP	79.1	27.78
YOLOv4	77.7	25.48
YOLOv4-tiny	78.7	31.76
SPFPN-YOLOv4-tiny	79.6	38.00
YOLOv5s	75.3	44.44
YOLOv5m	73.3	23.58
改进YOLOv5	82.0	35.71

对比模型	YOLOv5	YOLOv5+ CBAM	YOLOv5+ EIoU	YOLOv5+ CBAM+EIoU
精确率/%	76.40	79.60	77.70	77.50
召回率/%	72.50	75.30	73.60	76.60
F1分数/%	75.00	77.00	76.00	77.00
mAP/%	75.30	80.10	77.60	82.00
帧速率 /frame·s^-1	44.44	35.84	43.48	35.71