作者：甘波^1,2, 邢伟寅^3,4, 刘洪义⁵, 梁姝², 蒲国林², 代超⁶

作者单位：1. 四川智视运通科技有限公司，四川 成都 610000;
2. 达州职业技术学院，四川 达州 635000;
3. 菲律宾国家大学，菲律宾 马尼拉 1101;
4. 绵阳职业技术学院，四川 绵阳 621000;
5. 重庆交通大学土木工程学院，重庆 400000;
6. 四川智慧高速科技有限公司，四川 成都 610000

关键词：深度学习;智慧交通;目标检测;YOLOv5s

摘要：

为快速准确识别道路凹坑，研究提出一种基于改进YOLOv5s的道路凹坑检测算法。在YOLOv5s算法的主干网络中结合高效通道注意力机制，提高对凹坑区域的关注度；然后在检测头使用高效解耦头，有利于对凹坑进行准确预测；同时在边框损失函数中增加归一化Wasserstein距离损失，提升对小目标的检测能力。改进后的算法对复杂路况的凹坑检测具有较高的精度，在Pothole Dataset扩展数据集上，mAP和Precision上均超过原算法。将算法用于智慧交通领域，以便能更快地修复道路上严重凹坑。

Road pothole detection algorithm based on improved YOLOv5s
GAN Bo^1,2, XING Weiyin^3,4, LIU Hongyi⁵, LIANG Shu², PU Guolin², DAI Chao⁶
1. Sichuan ZhiShiYunTong Technology Co., Ltd., Chengdu 610000, China;
2. Dazhou Vocational and Technical College, Dazhou 635000, China;
3. National University (Philippines), Manila 1101, Philippines;
4. Mianyang Polytechnic, Mianyang 621000, China;
5. School of Civil Engineering, Chongqing Jiaotong University, Chongqing 400000, China;
6. Sichuan Intelligent Expressway Technology Co., Ltd., Chengdu 610000, China
Abstract: To rapidly and accurately identify road potholes, a detection algorithm based on an improved YOLOv5s has been proposed. The algorithm incorporates an efficient channel attention mechanism into the backbone network of YOLOv5s to enhance focus on pothole areas; it then utilizes efficient decoupled head in the detection head, which is beneficial for accurate pothole prediction; additionally, the bounding box loss function is augmented with normalized Wasserstein distance loss to improve the detection capability for small targets. The improved algorithm demonstrates higher accuracy in pothole detection under complex road conditions, surpassing the original algorithm in both mAP and Precision on the expanded Pothole Dataset. Applying this algorithm in the field of intelligent transportation can facilitate the faster repair of severe potholes on roads.
Keywords: deep learning;intelligent transportation;object detection;YOLOv5s
2024, 50(4):160-165  收稿日期: 2023-08-10;收到修改稿日期: 2024-03-10
基金项目: 四川省科技计划重点研发项目(2022YFG0206)；四川省知识产权专项资金项目(2022-ZS-00156)；达州市“同心智库”(TXZK23D17)
作者简介: 甘波(1983-),男,四川眉山市人,副研究员,高级工程师,主要从事深度学习、软件架构方面的研究。
参考文献
[1] GIRSHICK R, DONAHUA J, DARRELL T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]//IEEE Conference on Computer Vision & Pattern Recognition, Columbus, Ohio, USA, 2014: 570-578.
[2] ROSS G. Fast R-CNN[J]. Computer Science, 2015, 9(1): 10-19.
[3] REN S, HE K, GIRSHICK R, et al. Faster R-CNN: Towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(6): 1137-1149.
[4] 彭伟康, 陈爱军, 吴东明, 等. 基于改进Faster R-CNN的水准泡缺陷检测方法[J]. 中国测试, 2021, 47(7): 6-12.
PENG W K, CHEN A J, WU D M, et al. Defect detection method of level bubble based on improved Faster R-CNN[J]. China Measurement & Test, 2021, 47(7): 6-12.
[5] REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: Unified, real-time object detection[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2016.
[6] REDMON J, FARHADI A. Yolo9000: better, faster, stronger[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017.
[7] REDMON J, FARHADI A. Yolov3: An incremental improvement[J/OL]. arXiv preprint arXiv: 1804.02767, 2018.[2023-08-05]. https://arxiv.org/abs/1804.02767.
[8] NEPAL U, ESLAMIAT H. Comparing YOLOv3, YOLOv4 and YOLOv5 for autonomous landing spot detection in faulty UAVs[J]. Sensors, 2022, 22(2): 464.
[9] ZHAO J, ZHANG X, YAN J, et al. A wheat spike detection method in UAV images based on improved YOLOv5[J]. Remote Sensing, 2021, 13(16): 3095.
[10] LIU W, ANGUELOV D, ERHAN D, et al. SSD: single shot multibox detector[C]//Proceedings of the 14th European Conference on Computervision. Berlin, Germany: Springer, 2016: 21-37.
[11] JOCHER G. YOLOv5 release v6.1[EB/OL]. (2022-02-22)[2023-08-05] https://github.com/ultralytics/yolov5/releases/tag/v6.1,2022.1,2,3,4,5,7.
[12] WANG Q L, WU B G, ZHU P F, et al. ECA-Net: Efficient channel attention for deep convolutional neural networks[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2020.
[13] LI C Y, LI L L, JIANG H L, et al. Yolov6: A single-stage object detection framework for industrial applications[J/OL]. arXiv preprint arXiv: 2209.02976, 2022.[2023-08-05]. https://ar5iv.labs.arxiv.org/html/2209.02976.
[14] WANG J W, XU C, YANG W, et al. A normalized Gaussian wasserstein distance for tiny object detection[J/OL]. arXiv preprint arXiv: 2110.13389, 2021.[2023-08-05]. https://arxiv.org/abs/2110.13389.
[15]  Annotated Potholes Image Dataset [EB/OL].[2023-08-05] https://public.roboflow.com/object-detection/pothole.