《电子技术应用》
您所在的位置:首页 > 模拟设计 > 设计应用 > 基于改进YOLOv5的路面裂缝检测方法
基于改进YOLOv5的路面裂缝检测方法
电子技术应用
王向前1,成高立1,胡鹏2,夏晓华2
1.陕西高速机械化工程有限公司,陕西 西安 710038;2.长安大学 公路养护装备国家工程研究中心,陕西 西安710064
摘要: 针对现有裂缝检测模型体积较大且检测精度不高的问题,提出一种基于轻量化网络的无人机航拍图像裂缝检测方法。首先,使用MobileNetv3网络替代YOLOv5的主干网络,降低模型大小;其次,引入C3TR和CBAM模块提高网络表征能力,将损失函数替换为EIOU以提高模型的鲁棒性。实验结果表明,该方法在自制数据集上获得98.9%的精度,相较于原始YOLOv5提高1.2%,模型大小减小51.5%,检测速度提高37%。改进后的模型在精度、大小和速度上均优于Faster-RCNN等4种常见裂缝检测模型,满足了裂缝检测的实时性、轻量化和精度需求。
中图分类号:TP391.41;U418.6 文献标志码:A DOI: 10.16157/j.issn.0258-7998.234577
中文引用格式: 王向前,成高立,胡鹏,等. 基于改进YOLOv5的路面裂缝检测方法[J]. 电子技术应用,2024,50(3):80-85.
英文引用格式: Wang Xiangqian,Cheng Gaoli,Hu Peng,et al. Pavement crack detection method based on improved YOLOv5[J]. Application of Electronic Technique,2024,50(3):80-85.
Pavement crack detection method based on improved YOLOv5
Wang Xiangqian1,Cheng Gaoli1,Hu Peng2,Xia Xiaohua2
1.Shanxi Expressway Mechanization Engineering Limited Company, Xi′an 710038, China; 2.National Engineering Research Center of Highway Maintenance Equipment, Chang′an University, Xi′an, 710064,China
Abstract: Aiming at the problem that the existing crack detection model is large in size and the detection accuracy is not high, this paper proposes a crack detection method for UAV aerial images based on lightweight network. Firstly, the MobileNetv3 network is used instead of the YOLOv5 backbone network to reduce the model size. Secondly, the C3TR and CBAM modules are introduced to improve the network characterization ability, and the loss function is replaced with EIOU to improve the robustness of the model. Experimental results show that the proposed method obtains 98.9% accuracy on the self-made dataset, which is 1.2% higher than the original YOLOv5, the model size is reduced by 51.5%, and the detection speed is increased by 37%. The improved model is superior to four common crack detection models such as Faster-RCNN in terms of accuracy, size and speed, which meets the real-time, lightweight and accuracy requirements of crack detection.
Key words : road surface crack detection;YOLOv5;object detection;C3TR;CBAM;EIOU

引言

近年来,我国公路蓬勃发展,公路保养维护任务贯穿路面整个使用阶段[1]。在裂缝出现初期及时实现病害检测并修复,可有效地减缓或防止初期裂缝的恶化,对于提高路面使用寿命、保障行车安全具有重要意义。

路面裂缝检测方法主要有3种:传统的人眼观察识别方法主观性强;常规图像处理方法存在开发成本大、检测精度不高等问题;卷积神经网络相较于常规图像处理方法具有泛化性好、开发成本低等优点,但存在模型体积较大、检测精度有待提高的问题。文献[2]通过实验表明R-CNN系列、SPP-net和SSD等现有卷积神经网络模型体积较大;文献[3]证明YOLO的参数量较上述目标检测算法较少。但YOLO[3-4]系列算法在实际应用中依然存在模型体积大、裂缝检测精度不高等问题[5]。


本文详细内容请下载:

https://www.chinaaet.com/resource/share/2000005920


作者信息:

王向前1,成高立1,胡鹏2,夏晓华2

1.陕西高速机械化工程有限公司  2.长安大学 公路养护装备国家工程研究中心


杂志订阅.jpg

此内容为AET网站原创,未经授权禁止转载。