[1]何凌志,王玉珏,周月娥,等.基于改进的YOLOv5算法路面检测设计[J].工业仪表与自动化装置,2023,(04):93-97.[doi:10.19950/j.cnki.cn61-1121/th.2023.04.017]
　HE Lingzhi,WANG Yujue,ZHOU Yuee,et al.Pavement detection design based on improved YOLOv5 algorithm[J].Industrial Instrumentation & Automation,2023,(04):93-97.[doi:10.19950/j.cnki.cn61-1121/th.2023.04.017]
点击复制

基于改进的YOLOv5算法路面检测设计

《工业仪表与自动化装置》[ISSN:1000-0682/CN:61-1121/TH]

卷:
期数:
2023年04期
页码:
93-97
栏目:
出版日期:
2023-08-15

文章信息/Info

Title:
Pavement detection design based on improved YOLOv5 algorithm
文章编号:
1000-0682（2023）03-0093-05
作者:
何凌志; 王玉珏; 周月娥; 周研逸
南京理工大学 紫金学院，江苏 南京，210023
Author(s):
HE Lingzhi;  WANG Yujue;  ZHOU Yuee; ZHOU Yanyi
Nanjing University of Science and Technology Zijin college,Jiangsu Nanjing 210023,China
关键词:
深度学习; YOLOv5算法; 路面检测; 网络剪枝; 稀疏化训练; 目标检测算法; 轻量化检测网络; 嵌入式部署
Keywords:
deep learning;  YOLOv5 algorithm;  road detection;  network pruning;  sparse training;  target detection algorithm;  lightweight detection network;  embedded deployment
分类号:
TN919
DOI:
10.19950/j.cnki.cn61-1121/th.2023.04.017
文献标志码:
A
摘要:
随着大规模的公路建设，公路路面检测对于已建成公路的维护保养尤为重要，但是目前的深度学习网络模型都较大，并且部署到嵌入式端会造成Al算力不足的问题。因此，智能算法检测渐渐进入人们视野，该文设计了基于改进的YOLOv5的路面检测算法的检测设计。提出了一种通过Network Slimming网络剪枝的方法对稀疏化训练的YOLOv5目标检测算法模型进行剪枝微调，并通过tensorboard网页观察BN缩放因子直方图变化从而确定剪枝微调的比例。经过实际测试对比，相较与正常训练的算法，通过稀疏化处理后剪枝训练的YOLOv5算法模型所占权重减小了6.5 MB，对路面坑洞检测中的平均准确率（mAP）达到了81.4%，相比原始YOLOv5算法提升了2.1%，同时出现漏检现象较少，具有较好的检测精度。
Abstract:
With the large-scale highway construction, highway pavement detection is particularly important for the maintenance of completed highways. However, the current deep learning network models are large, and deployment to the embedded terminal will cause the problem of insufficient Al computing power. Therefore, intelligent algorithm detection gradually enters people’s field of vision. This paper designs a detection design based on the improved YOLOv5 road detection algorithm. This paper proposes a method of pruning the sparsely trained YOLOv5 target detection algorithm model through Network Slimming network pruning, and observes the change of the BN scaling factor histogram through the tensorboard webpage to determine the proportion of pruning and fine-tuning. After the actual test comparison, compared with the normal training algorithm, the weight of the YOLOv5 algorithm model trained by pruning after sparse processing is reduced by 6.5MB, and the average accuracy rate (mAP) in road pothole detection reaches 81.4 %, which is 2.1% higher than the original YOLOv5 algorithm. At the same time, there are fewer missed detections and better detection accuracy.

参考文献/References:

[1]肖德琴，蔡家豪，林思聪，等．基于 IFSSD 卷积神经网络的柚子采摘目标检测模型[J].农业 机械学报，2020，51( 5) : 28－35，97.

[2]钱伍,王国中,李国平.改进YOLOv5 的交通灯实时检测鲁棒算法[J].计算机科学与探索,2022,16(1):231-241.
[3]杨亚峰,苏维均,秦勇,等.基于语义标签的高铁接触网图像目标检测研究[J]. 计算机仿真, 2020, 37(11): 146-149, 188.?/div>
[4]REDMON J , DIVVALA S , GIRSHICK R , et al. You only look once: unified, real-time object detection[C]// Computer Vision & Pattern Recognition. IEEE, 2016.?/div>
[5] Al-Masni M A, Al-Antari M A, Park J-M, et al. Simultaneous detection and classification of breast masses in digital mammograms via a deep learning YOLO-based CAD system[J]. Computer Methods and Programs in Biomedicine, 2018, 157: 85-94.
[6] Hendry, Chen R C. Automatic license plate recognition via sliding-window darknet-YOLO deep learning[J]. Image and Vision Computing, 2019, 87: 47-56.
[7]REDMON J, FARHADI A. Yolov3: An incremental improvement[J].arXiv preprint arXiv:1804.02767,2018.
[8] REDMON J , FARHADI A .YOLO9000: Better, Faster, Stronger[C]// IEEE Conference on Computer Vision & Pattern Recognition. IEEE, 2017:6517-6525.
[9] REDMON J , FARHADI A . YOLOv3: An Incremental Improvement［EB /OL］．2021-08-14．https://arxiv.org．/pdf/1804.02 767.pdf.
[10]REDMON J, FARHADI A. Yolov3: An incremental improvement[J]. arXiv preprint arXiv:1804.02767, 2018.?/div>
[11]万卓 , 叶明 , 刘凯 . 基于改进 YOLOv4 的电机端盖缺陷检测 [J]. 计算机系统应用 ,2021,30（03）:79-87.
[12]管军霖 , 智鑫 . 基于 YOLOv4 卷积神经网络的口罩佩戴检测方法 [J]. 现代信息科技 ,2020,4（11）:9-12.
[13]黄海新 , 金鑫 . 基于 YOLOv4 的小目标缺陷检测 [J]. 电子 世界 ,2021（05）:146-147.
[14]杨英彬 , 郭子彧 , 蔡利民 .YOLOv4 的车辆检测方法研究 [J]. 电子世界 ,2021（05）:80-81+87.
[15]赵燕姣,李 钢,姚琼辛,等基于改进YOLOv4算法在车辆检测中的应用[J].电子设计工程,2022,30(24):37-42.
[16]王莉, 何牧天, 徐硕, 等. 基于 YOLOv5s 网络的垃圾 分类和检测[J]. 包装工程, 2021, 42(8): 50-56.?div>[17]黄剑翔,朱硕.基于改进的YOLOv5算法道路目标检测分类技术研究[J].电子设计工程,2023,31(4):188-193.

相似文献/References:

[1]杨 帅,张有芬,李玉惠,等.基于深度卷积神经网络的车标分类[J].工业仪表与自动化装置,2017,(05):75.
　YANG Shuai,ZHANG Youfen,LI Yuhui,et al.Vehicle classification based on deep convolutional neural network[J].Industrial Instrumentation & Automation,2017,(04):75.
[2]徐先峰,黄刘洋,龚 美.基于卷积神经网络与双向长短时记忆网络组合模型的短时交通流预测[J].工业仪表与自动化装置,2020,(01):13.
　XU Xianfeng,HUANG Liuyang,GONG Mei.Short-term traffic flow prediction based on combined model of convolutional neural network and bidirectional long-term memory network[J].Industrial Instrumentation & Automation,2020,(04):13.
[3]张晓华,马 煜,杨晨辉,等.基于卷积神经网络的设备安装位置智能识别方法[J].工业仪表与自动化装置,2021,(01):13.[doi:10.3969/j.issn.1000-0682.2021.01.003]
　ZHANG Xiaohua,MA yu,YANG Chenhui,et al.Intelligent identification method of equipment installation position based on convolution neural network[J].Industrial Instrumentation & Automation,2021,(04):13.[doi:10.3969/j.issn.1000-0682.2021.01.003]
[4]苏 杨,卢 翔,李 琨,等.基于轻量深度学习网络的机房人物检测研究[J].工业仪表与自动化装置,2021,(01):100.[doi:10.3969/j.issn.1000-0682.2021.01.024]
　SU Yang,LU Xiang,LI Kun,et al.Research on computer room human detection based on lightweight deep learning network[J].Industrial Instrumentation & Automation,2021,(04):100.[doi:10.3969/j.issn.1000-0682.2021.01.024]
[5]卢 翔,苏 杨,余 萱,等.基于深度学习的机房人物重识别研究[J].工业仪表与自动化装置,2021,(02):104.[doi:10.19950/j.cnki.cn61-1121/th.2021.02.024]
　LU Xiang,SU Yang,YU Xuan,et al.Research on computer room character recognition based on deep learning[J].Industrial Instrumentation & Automation,2021,(04):104.[doi:10.19950/j.cnki.cn61-1121/th.2021.02.024]
[6]倪四清,左光恒,张 俊.高速公路建设远程视频监控系统的设计与实现[J].工业仪表与自动化装置,2022,(02):76.[doi:10.19950/j.cnki.cn61-1121/th.2022.02.016]
　NI Siqing,ZUO Guangheng,ZHANG Jun.Design and realization of remote video monitoring system for expressway construction[J].Industrial Instrumentation & Automation,2022,(04):76.[doi:10.19950/j.cnki.cn61-1121/th.2022.02.016]
[7]王怀志,高德欣.基于深度学习的矿井电力短期负荷预测方法[J].工业仪表与自动化装置,2024,(01):51.[doi:DOI:10.19950/j.cnki.CN61-1121/TH.2024.01.012]
　WANG Huaizhi,GAO Dexin.Mine power short-term load forecasting method based on deep learning[J].Industrial Instrumentation & Automation,2024,(04):51.[doi:DOI:10.19950/j.cnki.CN61-1121/TH.2024.01.012]
[8]骈璐璐,裴焕斗,张宇璇.多场景烟雾环境下改进的YOLOv5s烟雾检测算法[J].工业仪表与自动化装置,2024,(02):101.[doi:DOI:10.19950/j.cnki.CN61-1121/TH.2024.02.018]
　PIAN Lulu,PEI Huandou,ZHANG Yuxuan.Improved YOLOv5s smoke detection algorithm in multi-scenario smoke environment[J].Industrial Instrumentation & Automation,2024,(04):101.[doi:DOI:10.19950/j.cnki.CN61-1121/TH.2024.02.018]
[9]徐晓强,丁 峰,毕淑敏.基于高速通信的港口设备远程检测与控制技术研究[J].工业仪表与自动化装置,2024,(05):83.[doi:DOI:10.19950/j.cnki.cn61-1121/th.2024.05.016]
　XU Xiaoqiang,DING Feng,BI Shumin.Design of remote detection and control technology for port equipment based on high-speed mobile communication[J].Industrial Instrumentation & Automation,2024,(04):83.[doi:DOI:10.19950/j.cnki.cn61-1121/th.2024.05.016]
[10]张 柱,刘海龙.基于ResNet模型的智能物料分拣系统设计[J].工业仪表与自动化装置,2025,(03):20.[doi:10.19950/j.cnki.CN61-1121/TH.2025.03.004]
　ZHANG Zhu,LIU Hailong.Design of intelligent material sorting system based on ResNet[J].Industrial Instrumentation & Automation,2025,(04):20.[doi:10.19950/j.cnki.CN61-1121/TH.2025.03.004]

备注/Memo

备注/Memo:
收稿日期：2022-07-01

基金项目：
2021年江苏省大学生创新创业训练计划项目（202113654016Y）；
2022年江苏省大学生创新创业训练计划项目（202213654015Y）

第一作者：
何凌志(2001—)，男，汉，本科生，主要研究方向为深度学习技术。E-mail：1607803784@qq.com

通信作者：
王玉珏（1981—)，女，学士学位(在读硕士)，讲师，主要研究领域为自动化控制及物联网技术的研究。E-mail：41040781@qq.com
更新日期/Last Update: 1900-01-01