基于YOLOv5的轻量化桥梁缺陷识别

doi:10.13229/j.cnki.jdxbgxb.20250539

摘要/Abstract

摘要：

以YOLOv5算法为基础框架，通过引入FasterNet轻量化网络结构优化特征提取过程，降低模型计算复杂度；结合卷积块注意力模块（CBAM）注意力机制，提高模型对缺陷特征的关注度；采用SIoU损失函数改进边界框回归精度，提升定位准确性。实验结果表明：本文改进模型在自建桥梁缺陷数据集上取得了显著效果，精确度和mAP分别达到81.2%和71.5%，较基准模型有明显提升。本研究为桥梁缺陷的智能化检测提供了技术支撑，对推动桥梁养护管理的数字化转型具有重要意义。

关键词: 交通运输系统工程, 缺陷检测, 深度学习, 无人机视角

Abstract:

Based on the YOLOv5 algorithm， the feature extraction process was optimized by introducing the FasterNet lightweight network structure to reduce the computational complexity of the model. Combining Convolutional Block Attention Module （CBAM） attention mechanism to improve the model's attention to defect features. Adopting the SIoU loss function to improve the accuracy of bounding box regression and enhance localization accuracy. The experimental results show that the improved model has achieved significant results on the self built bridge defect dataset， with accuracy and mAP reaching 81.2% and 71.5%， respectively， which is a significant improvement compared to the benchmark model. This study provides technical support for the intelligent detection of bridge defects， which is of great significance for promoting the digital transformation of bridge maintenance management.

Key words: engineering of communications and transportation system, defect detection, deep learning, UAV perspective

中图分类号:

U447

王琳虹,刘宇阳,刘子昱,鹿应佳,张宇恒,黄桂树. 基于YOLOv5的轻量化桥梁缺陷识别[J]. 吉林大学学报(工学版), 2025, 55(9): 2958-2968.

Lin-hong WANG,Yu-yang LIU,Zi-yu LIU,Ying-jia LU,Yu-heng ZHANG,Gui-shu HUANG. Defect recognition of lightweight bridges based on YOLOv5[J]. Journal of Jilin University(Engineering and Technology Edition), 2025, 55(9): 2958-2968.

图/表 17

图1

图2

图3

图4

图5

图6

图7

图8

图9

图10

图11

图12

图13

表1

表2

表3

表4

参考文献 26

[1]	丁威, 夏哲, 舒江鹏, 等. 基于负压吸附爬壁机器人和Transformer的混凝土桥塔裂缝识别检测[J]. 中国公路学报, 2024, 37(2): 53-64.
	Ding Wei, Xia Zhe, Shu Jiang-peng, et al. Recognition and detection of concrete bridge tower cracks using a negative pressure adhesion wall-climbing robot and Transformer[J]. China Journal of Highway and Transport, 2024, 37(2): 53-64.
[2]	刘凯, 李家坡, 邓智允, 等. 无人机桥梁检测技术的经济效益分析[J]. 南昌工程学院学报, 2018(3): 51-53, 80.
	Liu Kai, Li Jia-po, Deng Zhi-yun, et al. Benefit analysis of unmanned aerial vehicle bridge detection technology[J]. Journal of Nanchang Institute of Technology, 2018(3): 51-53, 80.
[3]	邹金欢. 无人机在输电线路巡检中的应用及效益分析[J]. 中国战略新兴产业, 2023(36): 153-155.
	Zou Jin-huan. Application and benefit analysis of drones in transmission line inspection[J]. China Strategic Emerging Industry, 2023(36): 153-155.
[4]	Oppenheim D, Shani G, Erlich O, et al. Using deep learning for image-based potato tuber disease detection[J]. Phytopathology, 2019, 109(6): 1083-1087.
[5]	Wang J X, Su Y H, Yao J H, et al. Apple rapid recognition and processing method based on an improved version of YOLOv5[J]. Ecological Informatics, 2023, 77: No.102196.
[6]	贾雪莹, 赵春江, 周娟, 等. 基于改进YOLOv7模型的柑橘表面缺陷在线检测[J]. 农业工程学报, 2023, 39(23): 142-151.
	Jia Xue-ying, Zhao Chun-jiang, Zhou Juan, et al. Online detection of citrus surface defects using improved YOLOv7 modeling[J]. Transactions of the Chinese Society of Agricultural Engineering, 2023, 39(23): 142-151.
[7]	王庞伟. 基于金字塔特征融合的二阶段三维点云车辆检测[J]. 交通运输系统工程与信息, 2022, 22(5):107-116.
	Wang Pang-wei. Pyramid-feature-fusion-based two-stage vehicle detection via 3D point cloud[J]. Journal of Transportation Systems Engineering and Information Technology, 2022, 22(5): 107-116.
[8]	张河山, 范梦伟, 谭鑫, 等. 基于改进YOLOX的无人机航拍图像密集小目标车辆检测[J]. 吉林大学学报: 工学版, 2025, 55(4): 1307-1318.
	Zhang He-shan, Fan Meng-wei, Tan Xin, et al. Dense small target vehicle detection in UAV aerial images based on improved YOLOX[J]. Journal of Jilin University (Engineering and Technology Edition), 2025, 55(4): 1307-1318.
[9]	Yang Li-na, Chen Gang, Wen-yan Ci. Multiclass objects detection algorithm using DarkNet-53 and DenseNet for intelligent vehicles[J] EURASIP Journal on Advances in Signal Processing, 2023, 2023(1): 1-18.
[10]	Cao F, Xing B, Luo J C, et al. An efficient object detection algorithm based on improved YOLOv5 for High-Spatial-Resolution remote sensing images[J]. Remote Sensing, 2023, 15(15):No. 3755.
[11]	许晓阳, 高重阳. 改进YOLOv7-tiny的轻量级红外车辆目标检测算法[J]. 计算机工程与应用, 2024, 60(1): 74-83.
	Xu Xiao-yang, Gao Chong-yang. Improved YOLOv7-tiny lightweight infrared vehicle target detection algorithm[J]. Computer Engineering and Applications, 2024, 60(1):74-83.
[12]	彭卫兵, 张明见, 全柳萌, 等. 基于改进Mask R-CNN的钢桥多缺陷智能识别[J]. 中国公路学报, 2024, 37(2): 100-109.
	Peng Wei-bing, Zhang Ming-jian, Quan Liu-meng, et al. Intelligent recognition of multiple diseases in steel bridges based on improved Mask R-CNN[J]. China Journal of Highway and Transport, 2024, 37(2): 100-109.
[13]	陈昭俊. 二阶段锚框和类均衡损失的遥感图像目标检测[J]. 图学学报, 2023,44(2): 249-259.
	Chen Zhao-jun. Object detection in remote sensing image based on two-stage anchor and class balanced loss[J]. Journal of Graphics, 2023,44(2): 249-259.
[14]	蒋燕芳. 基于图像处理与深度学习的RC桥梁表观病害识别[D]. 重庆: 重庆大学土木工程学院, 2019.
	Jiang Yan-fang. Identification of apparent defects in RC bridge with image processing and deep learning[D]. Chongqing:College of Civil Engineering, Chongqing University, 2019.
[15]	Zeng G, Wu Z Z, Xu L P, et al. Efficient vision Transformer YOLOv5 for accurate and fast traffic fign detection[J]. Electronics, 2024, 13(5):No. 880.
[16]	徐鑫. 基于YOLOv5的医学图像病变区域检测研究与应用[D]. 镇江: 江苏大学计算机科学与通信工程学院, 2023.
	Xu Xin. Research and application of medical image lesion region detection based on YOLOv5[D]. Zhenjiang:School of Computer Science and Communication Engineering, Jiangsu University, 2023.
[17]	钟源, 袁家政, 李鸿天, 等. 改进YOLOv5的智慧课堂人脸检测算法[J] .计算机工程与应用, 2024, 60(11): 251-257.
	Zhong Yuan, Yuan Jia-zheng, Li Hong-tian, et al. Intelligent classroom face detection algorithm with improved YOLOv5[J]. Computer Engineering and Applications, 2024, 60(11): 251-257.
[18]	Rejin V, Sambath M. YOLOv8: a novel object detection algorithm with enhanced performance and robustness[C]∥International Conference on Advances in Data Engineering and Intelligent Computing Systems (ADICS), Chennai, India, 2024: 1-6.
[19]	Chen J R, Kao S H, He H, et al. Run, don't walk: chasing higher FLOPS for faster neural networks [C]∥IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR),Vancouver, Canada, 2023: 12021-12031.
[20]	Chen J, He T, Zhuo W, et al. TVConv: efficient translation variant convolution for layout-aware visual processing [C]∥IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), New Orleans, USA, 2022: 12538-12548.
[21]	张楠楠, 张晓, 白铁成, 等. 基于CBAM-YOLOv7的自然环境下棉叶病虫害识别方法[J]. 农业机械学报, 2023, 54(201): 239-244.
	Zhang Nan-nan, Zhang Xiao, Bai Tie-cheng, et al. Pest and disease identification method for cotton leaves in natural environments based on CBAM-YOLOv7[J]. Transactions of the Chinese Society for Agricultural Machinery, 2023, 54(201): 239-244.
[22]	Woo S, Park J, Lee J Y, et al. CBAM: convolutional block attention module[J]. Lecture Notes in Computer Science (Including Subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), 2018, 11211: 3-19.
[23]	Hamid R, Nathan T, Junyoung G, et al. Generalized intersection over union: a metric and a loss for bounding box regression[C]∥IEEE/CVF Conference on Computer Vision and Pattern Recognition, Long Beach, USA, 2019: 658-666.
[24]	Gevorgyan Z. SIoU Loss: more powerful learning for bounding box regression[DB/OL]. [2025-06-05]. .
[25]	王一旭, 肖小玲, 王鹏飞, 等. 改进YOLOv5s的小目标烟雾火焰检测算法[J]. 计算机工程与应用, 2023, 59(1): 72-81.
	Wang Yi-xu, Xiao Xiao-ling, Wang Peng-fei, et al. Improved YOLOv5s small target smoke and fire detection algorithm[J]. Computer Engineering and Applications, 2023, 59(1): 72-81.
[26]	孙长亮, 何峻, 肖怀铁. 基于ROC曲线的目标识别性能评估方法[J]. 雷达科学与技术, 2007, 5(1): 17-21.
	Sun Chang-liang, He Jun, Xiao Huai-tie. A new performance evaluation method based on ROC curve[J]. Radar Science and Technology, 2007, 5(1): 17-21.

相关文章 15

[1]	曲昭伟,王铭阳,王喆,宋现敏,张云翔,黄镜尘. 基于自动驾驶模块化车辆主辅功能分配的公交自适应调度方法[J]. 吉林大学学报(工学版), 2025, 55(9): 2946-2957.
[2]	穆长儒,徐亮,程国柱. 基于能量合理分配的外包U型钢-混凝土组合护栏防撞性能[J]. 吉林大学学报(工学版), 2025, 55(8): 2669-2680.
[3]	廉敬,张继保,刘冀钊,张家骏,董子龙. 基于文本引导的人脸图像修复[J]. 吉林大学学报(工学版), 2025, 55(8): 2732-2740.
[4]	刘元宁,王星喆,黄子彧,张家晨,刘震. 基于多模态数据融合的胃癌患者生存预测模型[J]. 吉林大学学报(工学版), 2025, 55(8): 2693-2702.
[5]	李艳波,汪静远,陈圆媛,程绍峰,吕浩楠,陈俊硕. 面向高速公路服务区自洽能源系统的RAMS评价方法[J]. 吉林大学学报(工学版), 2025, 55(7): 2243-2250.
[6]	柴树山,周志强,李海涛,徐炅旸. 基于图时空模式学习网络的路网实时交通事件自动检测方法[J]. 吉林大学学报(工学版), 2025, 55(7): 2145-2161.
[7]	于江波,翁剑成,林鹏飞,孙宇星,柴娇龙. 基于混合Transformer的对外客运枢纽抵站客流预测模型[J]. 吉林大学学报(工学版), 2025, 55(7): 2251-2259.
[8]	戢晓峰,邓若凡,乔新,关昊天. 建成环境对共享单车时间集聚模式的非线性影响[J]. 吉林大学学报(工学版), 2025, 55(7): 2233-2242.
[9]	袁靖舒,李武,赵兴雨,袁满. 基于BERTGAT-Contrastive的语义匹配模型[J]. 吉林大学学报(工学版), 2025, 55(7): 2383-2392.
[10]	闫晟煜,程铭杰,田宏策,王洪瑀,周永恒,马博浩. 封闭式景区纯电动客车调度方法[J]. 吉林大学学报(工学版), 2025, 55(6): 1984-1993.
[11]	赵红专,吴泽健,张鑫,石胜文,李文勇,展新,许恩永,王佳明. 基于密度离散度和信息传输延迟的网联商用车弯道格子模型[J]. 吉林大学学报(工学版), 2025, 55(6): 2015-2029.
[12]	潘福全,牛远征,张丽霞,杨金顺,陈秀锋,陈德启. 智能网联环境下无信号交叉口车辆通行控制策略[J]. 吉林大学学报(工学版), 2025, 55(6): 1948-1962.
[13]	徐慧智,郝东升,徐小婷,蒋时森. 基于深度学习的高速公路小目标检测算法[J]. 吉林大学学报(工学版), 2025, 55(6): 2003-2014.
[14]	张汝波,常世淇,张天一. 基于深度学习的图像信息隐藏方法综述[J]. 吉林大学学报(工学版), 2025, 55(5): 1497-1515.
[15]	潘义勇,徐家聪,尤逸文,全勇俊. 网约车出行需求影响因素多尺度空间异质性分析[J]. 吉林大学学报(工学版), 2025, 55(5): 1567-1575.

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed

超参数	数值	超参数	数值
初始学习率	0.012 40	明度增益	0.375 0
学习率因子	0.011 70	旋转角度/（°）	0.000 0
动量	0.980 00	平移幅度	0.090 1
权重衰减	0.000 48	缩放幅度	0.615 0
预热轮次	4.080 00	剪切幅度	0.000 0
预热动量	0.711 00	透视变换	0.000 0
预热偏置学习率	0.108 00	上下翻转概率	0.000 0
边界框损失权重	0.048 30	左右翻转概率	0.500 0
类别损失权重	0.378 00	马赛克增强概率	0.825 0
类别损失权重因子	1.020 00	混合增强概率	0.000 0
目标存在损失权重	0.976 00	复制粘贴增强概率	0.000 0
目标存在损失权重因子	1.070 00	锚框尺度系数	3.690 0
IoU阈值（训练）	0.200 00	色调增益	0.012 4
锚框匹配阈值因子	2.920 00	饱和度增益	0.792 0
Focal Loss 伽马值	0.000 00

CBAM	FasterNet	SIoU	精确率	召回率	平均检测精度	F₁分数
			0.782	0.670	0.699	0.722
√			0.797	0.675	0.705	0.731
√	√		0.806	0.660	0.711	0.726
√	√	√	0.812	0.686	0.715	0.744

模型	精确率	召回率	平均检测精度	参数量	帧率（帧·s^-1）
YOLOv5n	0.683	0.634	0.651	205 728 8	34.44
YOLOv5s	0.781	0.670	0.698	702 772 0	39.84
YOLOv8s	0.731	0.661	0.683	605 728 8	53.01
改进后的YOLOv5	0.812	0.686	0.715	635 776 0	52.08

模型	平均检测精度	精确率	召回率	帧率/（帧·s^-1）
Faster R-CNN	0.507	0.748	0.531	13.62
SSD	0.512	0.676	0.542	35.62
YOLOv5s	0.698	0.781	0.670	39.84
YOLOv8s	0.683	0.731	0.661	53.01
YOLOv11	0.696	0.793	0.686	55.24
改进后的YOLOv5	0.715	0.812	0.686	52.08