一种基于改进YOLOv8的轻量化路面病害检测算法

doi:10.11996/JG.j.2095-302X.2024050892

图学学报 ›› 2024, Vol. 45 ›› Issue (5): 892-900.DOI: 10.11996/JG.j.2095-302X.2024050892

• 图像处理与计算机视觉 • 上一篇下一篇

一种基于改进YOLOv8的轻量化路面病害检测算法

胡凤阔¹(), 叶兰¹(), 谭显峰², 张钦展³, 胡志新¹, 方清¹, 王磊², 满孝锋³

1.东华理工大学机械与电子工程学院，江西南昌 330013
2.江西省交通工程质量监督站试验检测中心，江西南昌 330006
3.中化学交通建设集团第二工程有限公司，山东青岛 266000

收稿日期:2024-05-01 修回日期:2024-07-10 出版日期:2024-10-31 发布日期:2024-10-31
通讯作者:叶兰(1987-)，女，副教授，博士。主要研究方向为机器视觉感知与控制技术。E-mail：yelan87@163.com
第一作者:胡凤阔(2000-)，男，硕士研究生。主要研究方向为计算机视觉和目标检测。E-mail：1754494774@qq.com
基金资助:
江西省交通运输厅科技项目(2023H0031);博士科研启动基金项目(DHBK2023007)

A refined YOLOv8-based algorithm for lightweight pavement disease detection

HU Fengkuo¹(), YE Lan¹(), TAN Xianfeng², ZHANG Qinzhan³, HU Zhixin¹, FANG Qing¹, WANG Lei², MAN Xiaofeng³

1. School of Mechanical and Electronic Engineering, East China University of Technology, Nanchang Jiangxi 330013, China
2. Jiangxi Traffic Engineering Quality Supervision Station Test Center, Nanchang Jiangxi 330006, China
3. China National Chemical Communications Construction Group Second Engineering Co. LTD, Qingdao Shandong 266000, China

Received:2024-05-01 Revised:2024-07-10 Published:2024-10-31 Online:2024-10-31
Contact: YE Lan (1987-), associate professor, Ph.D. Her main research interests cover machine vision perception and control technology. E-mail：yelan87@163.com
First author：HU Fengkuo (2000-), master student. His main research interests cover computer vision and target detection. E-mail：1754494774@qq.com
Supported by:
Jiangxi Provincial Department of Transportation Science and Technology Project(2023H0031);Project of Doctoral Research Initiation Fund(DHBK2023007)

摘要/Abstract

摘要：

路面病害检测是实现道路损伤修复、确保行车安全的关键任务。针对现有路面病害检测算法精度低、成本高、模型参数大及难以应用于移动终端设备等问题，提出了一种基于改进YOLOv8n模型的轻量级检测算法YOLOv8n-GSBP。首先，通过在骨干网络引入C2f-GhostNetv2模块保证检测精度并实现了模型轻量化，同时在SPPF模块后加入SimAM注意力机制模块，增强了网络对路面病害特征提取与背景环境特征区分的能力；其次，通过在颈部网络更换BiFPN结构增强模型多尺度特征融合能力，提升精确度和鲁棒性的同时解决了路面病害尺度差异较大问题；最后，基于参数共享原理改进检测头，并引入空间通道重建卷积模块SCConv，实现了检测头的轻量化，降低了模型参数和计算量。在RDD2022数据集上的实验结果表明，YOLOv8n-GSBP路面病害检测方法相较于YOLOv8n网络mAP50虽只提高了0.3%，但参数量降低了55.6%、计算量大幅度降低至36.7%，实现了对道路病害的实时准确检测。通过与其他主流目标检测算法的对比，进一步验证了算法的有效性和优越性。

关键词: 深度学习, 路面病害检测, YOLOv8n, 注意力机制, 轻量级算法

Abstract:

Road surface defect detection is a crucial task for repairing road damage and ensuring driving safety. To address the issues of low detection accuracy, high costs, large model parameters, and the difficulty in applying existing road surface defect detection algorithms to mobile terminal devices, a lightweight detection algorithm, YOLOv8n-GSBP, based on the improved YOLOv8n model, was proposed. Firstly, the C2f-GhostNetv2 module was introduced into the backbone network to maintain detection accuracy while achieving model lightweight. Additionally, the SimAM module was added after the SPPF module to enhance the network’s ability to extract road surface defect features and distinguish them from background environmental features. Secondly, the neck network was replaced with the BiFPN structure, and the model’s multi-scale feature fusion capability was enhanced while addressing significant differences in road surface defect scales to improve precision and robustness. Finally, the head was improved by the parameter-sharing principle, and the spatial channel reconstruction convolutional module SCConv was introduced to achieve lightweight improvement of the detection head while reducing model parameters and computational complexity. The experimental results on the RDD2022 dataset showed that the mAP50 of YOLOv8n-GSBP road surface disease detection method was 0.3% higher than that of the YOLOv8n; however, the parameters were reduced by 55.6% and the computational complexity was reduced to 36.7%. Furthermore, through comparisons with other mainstream object detection algorithms, we further validated both effectiveness and superiority of our proposed algorithm.

Key words: deep learning, pavement disease detection, YOLOv8n, attention mechanism, lightweight algorithm

中图分类号:

胡凤阔, 叶兰, 谭显峰, 张钦展, 胡志新, 方清, 王磊, 满孝锋. 一种基于改进YOLOv8的轻量化路面病害检测算法[J]. 图学学报, 2024, 45(5): 892-900.

HU Fengkuo, YE Lan, TAN Xianfeng, ZHANG Qinzhan, HU Zhixin, FANG Qing, WANG Lei, MAN Xiaofeng. A refined YOLOv8-based algorithm for lightweight pavement disease detection[J]. Journal of Graphics, 2024, 45(5): 892-900.

图/表 13

图1 YOLOv8n-GSBP网络结构图

Fig. 1 Network structure diagram of YOLOv8n-GSBP

图2 C2f-GhostNetv2模块结构

Fig. 2 C2f-GhostNetv2 module structure

图3 GhostNetv2瓶颈模块结构

Fig. 3 GhostNetv2 bottleneck module architecture

图4 SimAM注意力机制结构图

Fig. 4 Sim attention mechanism structure diagram

图5 特征融合网络结构对比图((a) PANet结构[16]；(b) BiFPN结构[17])

Fig. 5 Comparison of feature fusion network structures ((a) PANet structure[16]; (b) BiFPN structure[17])

图6 检测头整体结构图

Fig. 6 Overall structure of the detection head

表1 实验环境及参数配置

Table 1 Experimental environment and parameter configuration

名称	参数
操作系统	Windows 10
CPU	13th Gen Intel(R) Core(TM) i5-13400F
显卡	Nvidia RTX3060
显存	12 GB
编程语言	Python3.8
深度学习框架	Pytorch 1.12.1+CUDA 11.3.1
优化函数	SGD
训练轮数	300
初始学习率	0.01
批量	16

表2 不同注意力机制对比结果

Table 2 Comparative results of different attention mechanisms

Algorithms	mAP50/%	mAP50:90/%	F1	GFLOPs/G	Params/M
YOLOv8n	57.0	29.4	57.8	8.1	3.0
YOLOv8n+BLRA	56.7	28.5	57.6	8.3	3.42
YOLOv8n+TA	57.2	28.6	57.9	8.2	3.2
YOLOv8n+CA	57.4	28.9	58.3	8.1	3.0
YOLOv8n+SimAM	57.5	29.1	58.4	8.1	3.0

表3 检测头中不同卷积模块检测对比结果

Table 3 Comparison results of detection of different convolutional modules in the detection header

Algorithms	mAP50/%	mAP50:90/%	F1	GFLOPs/G	Params/M
YOLOv8n	57.0	29.4	57.8	8.1	3.0
v8n+RepConv	57.3	29.6	58.1	8.5	3.1
v8n+EMSConv	56.6	29.2	57.3	6.0	2.6
v8n+SCConv	56.8	29.3	57.6	5.8	2.6

图7 原YOLOv8n检测结果((a)增加GhostNetv2+SimAM模块前；(b)采用BiFPN结构前；(c)采用PSS检测头前；(d) 4种模块未改进前)

Fig. 7 Original YOLOv8n detection results ((a) Before adding GhostNetv2+SimAM module; (b) Before using the BiFPN; (c) Before using the PSS detection head; (d) None of the four modules had been improved)

图8 本文YOLOv8n-GSBP算法检测结果((a)增加GhostNetv2+SimAM模块后；(b)采用BiFPN结构后；(c)采用PSS检测头后；(d) 4种模块改进后)

Fig. 8 Detection results of the YOLOv8n-GSBP algorithm proposed in this paper ((a) After adding GhostNetv2+SimAM module; (b) After using the BiFPN; (c) After using the PSS detection head; (d) Four modules are improved)

表4 消融实验对比结果

Table 4 Comparative results of ablation experiments

Algorithms	GhostNetv2+SimAM	BiFPN	PSS	mAP50/%	mAP50:90/%	F1	GFLOPs/G	Params/M	FPS
YOLOv8n				57.0	29.4	57.8	8.1	3.0	152.9
YOLOv8n-GS	√			57.5	29.1	58.3	6.8	2.6	147.4
YOLOv8n-B		√		57.1	29.9	57.9	7.1	2.0	132.8
YOLOv8n-P			√	56.8	29.3	57.6	5.8	2.6	145.0
YOLOv8n-GSB	√	√		57.5	29.4	58.2	5.8	1.6	123.5
YOLOv8n-GSP	√		√	57.2	29.1	58.1	4.5	2.2	134.7
YOLOv8n-BP		√	√	56.9	29.4	57.7	4.8	1.6	122.1
YOLOv8n-GSBP	√	√	√	57.3	29.5	58.1	3.6	1.1	110.0

表5 不同算法对比结果

Table 5 Comparison results of different algorithms

Algorithms	mAP50/%	mAP50:90/%	F1	GFLOPs/G	Params/M
FasterR-CNN	51.2	22.5	49.4	370.2	137.1
Detr	54.3	30.4	55.0	101.0	31.2
YOLOv5n	54.1	28.7	54.8	7.1	2.5
YOLOv6n	55.1	27.1	56.0	11.2	4.2
YOLOv7n-tiny	56.5	25.7	57.1	13.0	6.9
Damo-YOLO	56.9	27.6	57.6	6.2	5.5
YOLOv8n	57.0	29.4	57.8	8.1	3.0
YOLOv10n	55.5	28.9	56.3	8.2	2.7
YOLOv8n-GSBP	57.3	29.5	58.1	3.6	1.1

参考文献 22

[1]	宋平丽. 基于视频图像的桥梁裂缝检测[D]. 武汉: 武汉理工大学, 2010.
	SONG P L. Crack detection of bridge on video image[D]. Wuhan: Wuhan University of Technology, 2010 (in Chinese).
[2]	周林. 基于图像处理的路面裂缝检测系统设计与研究[D]. 太原: 太原理工大学, 2013.
	ZHOU L. Research and design of pavement crack detection system based on image processing[D]. Taiyuan: Taiyuan University of Technology, 2013 (in Chinese).
[3]	孙朝云, 裴莉莉, 李伟, 等. 基于改进Faster R-CNN的路面灌封裂缝检测方法[J]. 华南理工大学学报: 自然科学版, 2020, 48(2): 84-93.
	SUN Z Y, PEI L L, LI W, et al. Pavement sealed crack detection method based on improved faster R-CNN[J]. Journal of South China University of Technology: Natural Science Edition, 2020, 48(2): 84-93 (in Chinese).
[4]	刘凡, 王君锋, 陈峙宇, 等. 基于并行注意力UNet的裂缝检测方法[J]. 计算机研究与发展, 2021, 58(8): 1718-1726.
	LIU F, WANG J F, CHEN Z Y, et al. Parallel attention based UNet for crack detection[J]. Journal of Computer Research and Development, 2021, 58(8): 1718-1726 (in Chinese).
[5]	张政超. 改进YOLOv5的轻量级带钢表面缺陷检测[J]. 计算机系统应用, 2023, 32(6): 278-285.
	ZHANG Z C. Lightweight strip steel defect detection based on improved YOLOv5[J]. Computer Systems & Applications, 2023, 32(6): 278-285 (in Chinese).
[6]	蒋博, 万毅, 谢显中. 改进YOLOv5s的轻量化钢材表面缺陷检测模型[J]. 计算机科学, 2023, 50(S2): 230900113.
	JIANG B, WAN Y, XIE X Z. Improved YOLOv5s lightweight steel surface defect detection model[J]. Computer Science, 2023, 50(S2): 230900113 (in Chinese).
[7]	贾玉进, 张振程, 李浠铭, 等. 基于轻量化YOLOv5网络的输电线路绝缘子缺陷检测[J]. 电力学报, 2024, 39(1): 36-44.
	JIA Y J, ZHANG Z C, LI X M, et al. Defect detection of insulator on transmission line based on lightweight YOLOv5 network[J]. Journal of Electric Power, 2024, 39(1): 36-44 (in Chinese).
[8]	HOWARD A G, SANDLER M, CHU G, et al. Searching for MobileNetV3[C]// 2019 IEEE/CVF International Conference on Computer Vision. New York: IEEE Press, 1314-1324.
[9]	ZHANG 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. New York: IEEE Press, 2018: 6848-6856.
[10]	HAN K, WANG Y H, TIAN Q, et al. GhostNet: more features from cheap operations[C]// 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2020: 1577-1586.
[11]	马燕婷, 赵红东, 阎超, 等. 改进YOLOv5网络的带钢表面缺陷检测方法[J]. 电子测量与仪器学报, 2022, 36(8): 150-157.
	MA Y T, ZHAO H D, YAN C, et al. Strip steel surface defect detection method by improved YOLOv5 network[J]. Journal of Electronic Measurement and Instrumentation, 2022, 36(8): 150-157 (in Chinese).
[12]	魏陈浩, 杨睿, 刘振丙, 等. 具有双层路由注意力的YOLOv8道路场景目标检测方法[J]. 图学学报, 2023, 44(6): 1104-1111. DOI
	WEI C H, YANG R, LIU Z B, et al. YOLOv8 with bi-level routing attention for road scene object detection[J]. Journal of Graphics, 2023, 44(6): 1104-1111 (in Chinese).
[13]	耿焕同, 刘振宇, 蒋骏, 等. 基于改进YOLOv8的嵌入式道路裂缝检测算法[J]. 计算机应用, 2024, 44(5): 1613-1618. DOI
	GENG H T, LIU Z Y, JIANG J, et al. Embedded road crack detection algorithm based on improved YOLOv8[J]. Journal of Computer Applications, 2024, 44(5): 1613-1618 (in Chinese). DOI
[14]	崔克彬, 焦静颐. 基于MCB-FAH-YOLOv8的钢材表面缺陷检测算法[J]. 图学学报, 2024, 45(1): 112-125. DOI
	CUI K B, JIAO J Y. Steel surface defect detection algorithm based on MCB-FAH-YOLOv8[J]. Journal of Graphics, 2024, 45(1): 112-125 (in Chinese). DOI
[15]	YANG L X, ZHANG R Y, LI L D, et al. SimAM:a simple, parameter-free attention module for convolutional neural networks[C]// The 38th International Conference on Machine Learning. New York: PLMR, 2021: 11863-11874.
[16]	WANG K X, LIEW J H, ZOU Y T, et al. PANet: few-shot image semantic segmentation with prototype alignment[C]// 2019 IEEE/CVF International Conference on Computer Vision. New York: IEEE Press, 2019: 9196-9205.
[17]	TAN 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. New York: IEEE Press, 2020: 10778-10787.
[18]	LI J F, WEN Y, HE L H. SCConv: spatial and channel reconstruction convolution for feature redundancy[C]// 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2023: 6153-6162.
[19]	ARYA D, MAEDA H, GHOSH S K, et al. RDD2022: a multi-national image dataset for automatic Road Damage Detetion[EB/OL]. [2024-01-11]. https://arxiv.org/abs/2209.08538.
[20]	ZHU L, WANG X J, KE Z H, et al. BiFormer: vision transformer with bi-level routing attention[C]// 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2023: 10323-10333.
[21]	MISRA D, NALAMADA T, ARASANIPALAI A U, et al. Rotate to attend: convolutional triplet attention module[C]// 2021 IEEE Winter Conference on Applications of Computer Vision. New York: IEEE Press, 2021: 3138-3147.
[22]	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.

一种基于改进YOLOv8的轻量化路面病害检测算法

A refined YOLOv8-based algorithm for lightweight pavement disease detection

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 13

参考文献 22

相关文章 15

编辑推荐

Metrics

本文评价

[1]	李琼 , 考月英 , 张莹 , 徐沛 . 面向无人机航拍图像的目标检测研究综述[J]. 图学学报, 2024, 45(6): 1145-1164.
[2]	刘灿锋, 孙浩, 东辉. 结合 Transformer 与 Kolmogorov Arnold 网络的分子扩增时序预测研究[J]. 图学学报, 2024, 45(6): 1256-1265.
[3]	宋思程, 陈辰, 李晨辉, 王长波. 基于密度图多目标追踪的时空数据可视化[J]. 图学学报, 2024, 45(6): 1289-1300.
[4]	李珍峰, 符世琛, 徐乐, 孟博, 张昕, 秦建军. 基于MBI-YOLOv8的煤矸石目标检测算法研究[J]. 图学学报, 2024, 45(6): 1301-1312.
[5]	王宗继, 刘云飞, 陆峰. Cloud Sphere: 一种基于渐进式变形自编码的三维模型表征方法[J]. 图学学报, 2024, 45(6): 1375-1388.
[6]	许丹丹, 崔勇, 张世倩, 刘雨聪, 林予松. 优化医学影像三维渲染可视化效果：技术综述[J]. 图学学报, 2024, 45(5): 879-891.
[7]	刘义艳, 郝婷楠, 贺晨, 常英杰. 基于DBBR-YOLO的光伏电池表面缺陷检测[J]. 图学学报, 2024, 45(5): 913-921.
[8]	吴沛宸, 袁立宁, 胡皓, 刘钊, 郭放. 基于注意力特征融合的视频异常行为检测[J]. 图学学报, 2024, 45(5): 922-929.
[9]	翟永杰, 李佳蔚, 陈年昊, 王乾铭, 王新颖. 融合改进Transformer的车辆部件检测方法[J]. 图学学报, 2024, 45(5): 930-940.
[10]	刘丽, 张起凡, 白宇昂, 黄凯烨. 结合Swin Transformer的多尺度遥感图像变化检测研究[J]. 图学学报, 2024, 45(5): 941-956.
[11]	姜晓恒, 段金忠, 卢洋, 崔丽莎, 徐明亮. 融合先验知识推理的表面缺陷检测[J]. 图学学报, 2024, 45(5): 957-967.
[12]	章东平, 魏杨悦, 何数技, 徐云超, 胡海苗, 黄文君. 特征融合与层间传递：一种基于Anchor DETR改进的目标检测方法[J]. 图学学报, 2024, 45(5): 968-978.
[13]	谢国波, 林松泽, 林志毅, 吴陈锋, 梁立辉. 基于改进YOLOv7-tiny的道路病害检测算法[J]. 图学学报, 2024, 45(5): 987-997.
[14]	熊超, 王云艳, 罗雨浩. 特征对齐与上下文引导的多视图三维重建[J]. 图学学报, 2024, 45(5): 1008-1016.
[15]	彭文, 林金炜. 基于空间信息关注和纹理增强的短小染色体分类方法[J]. 图学学报, 2024, 45(5): 1017-1029.