[1]吴裕发,郑少峰.基于YOLOv8改进的下水管道障碍物识别算法[J].福建理工大学学报,2024,22(06):590-597.[doi:10.3969/j.issn.2097-3853.2024.06.012]
　WU Yufa,ZHENG Shaofeng.Identification algorithm of sewer obstruction based on YOLOv8[J].Journal of Fujian University of Technology;,2024,22(06):590-597.[doi:10.3969/j.issn.2097-3853.2024.06.012]
点击复制

基于YOLOv8改进的下水管道障碍物识别算法
分享到：

《福建理工大学学报》[ISSN:2097-3853/CN:35-1351/Z]

卷:
第22卷
期数:
2024年06期
页码:
590-597
栏目:
出版日期:
2024-12-25

文章信息/Info

Title:
Identification algorithm of sewer obstruction based on YOLOv8
作者:
吴裕发; 郑少峰
福建理工大学交通运输学院
Author(s):
WU Yufa;  ZHENG Shaofeng
School of Transportation, Fujian University of Technology
关键词:
目标检测; 下水管道; 障碍物识别; PGI; SCConv; Focal-Modulation
Keywords:
target detection; sewer pipe; obstacle detection; PGI; SCConv; Focal-Modulation
分类号:
TP391.4
DOI:
10.3969/j.issn.2097-3853.2024.06.012
文献标志码:
A
摘要:
为提升下水管道障碍物清理效率和管道障碍物识别准确率，提出一种基于YOLOv8改进的管道障碍物识别算法。通过优化YOLOv8目标检测模型，使其更适用于环境复杂的管道内部障碍物检测任务。在YOLOv8网络结构的基础上引入PGI模块，增加辅助可逆支路和多级辅助模块缓解信息瓶颈问题，减少精度损失；引入SCConv模块来替换C2f模块，在实现模型轻量化的情况下保持检测精度；引入Focal?Modulation模块改进了传统的SPPF模块，使模型的精度得到一定程度的提升。实验结果表明，改进后的识别算法与YOLOv8n模型相比，在mAP@0.5精度上提升4.6%，在mAP@0.5~0.95精度上提升3.9%，参数量降低33.3%，计算量减少17.3%，更加适用于下水管道障碍物的识别检测。
Abstract:
In order to improve the cleaning efficiency of sewer pipe obstacles and the accuracy of pipeline obstacle recognition, an improved pipeline obstacle recognition algorithm based on YOLOv8 is proposed. By optimizing the YOLOv8 target detection model, the algorithm is made more suitable for obstacle detection in the complex environment inside the pipeline. Based on the network structure of YOLOv8, the PGI module proposed is introduced. The addition of auxiliary reversible branches and multistage auxiliary module in the module effectively alleviates the information bottleneck problem, as a result significantly reducing the loss of accuracy. The SCConv module is introduced to replace the C2f module to maintain the detection accuracy while realizing the lightweight of the model. The introduction of Focal-Modulation module improves the traditional SPPF module, so that the accuracy of the model is improved to a certain extent. Experimental results show that compared with YOLOv8n model, the improved recognition algorithm improves the accuracy of mAP@0.5 by 4.6%, improves the accuracy of mAP @0.5 ～0.95 by 3.9%, reduces the number of parameters by 33.3%, and reduces the amount of calculation by 17.3%. It is more suitable for the identification and detection of obstacles in sewer pipes.

参考文献/References:

［1］ LI B Z，JIANG W Z，GU J J. Research on target detection algorithm based on deep learning technology［C］∥ IEEE International Conference on Power Electronics，Computer Applications. Shenyang: IEEE，2021：137-142. ［2］ 郑远攀，李广阳，李晔. 深度学习在图像识别中的应用研究综述［J］. 计算机工程与应用，2019，55（12）：20-36..［3］ WEI S，RAMAKRISHNA V，KANADE T，et al. Convolutional pose machines［C］∥IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas，NV: IEEE，2016：4724-4732. ［4］ ZHANG C L，SHI L，BU N，et al. Automatic correction method of industrial instrument images based on YOLOv8 keypoint detection and perspective transformation［M］∥Lecture Notes in Computer Science. Singapore：Springer, 2024：3-11. ［5］ CAO J ，ZHANG T ，HOU L ，et al. An improved YOLOv8 algorithm for small object detection in autonomous driving［J］. Journal of Real-Time Image Processing， 2024，21(4):1-16. ［6］ SAMPATH S，CHAURASIYA K L，ARYAN P，et al. An innovative approach towards defect detection and localization in gas pipelines using integrated in-line inspection methods［J］. Journal of Natural Gas Science and Engineering，2021，90：103933. ［7］ 邵延华，张铎，楚红雨，等. 基于深度学习的YOLO目标检测综述［J］. 电子与信息学报，2022，44（10）：3697-3708. ［8］ WANG C Y，YEH I H，LIAO H Y M. YOLOv9：learning what you want to learn using programmable gradient information［EB/OL］. 2024：2402.13616.https：∥arxiv.org/abs/2402.13616v2.［9］ LI J F，WEN Y，HE L H. SCConv：spatial and channel reconstruction convolution for feature redundancy［C］∥ IEEE/CVF Conference on Computer Vision and Pattern Recognition. Vancouver，BC: IEEE，2023：6153-6162. ［10］ YANG J W，LI C Y，DAI X Y，et al. Focal modulation networks［C］∥Proceedings of the 36th International Conference on Neural Information Processing Systems. New Orleans: ACM, 2022: 4203-4217. ［11］ HAURUM J B，MOESLUND T B. SewerML：a multi-label sewer defect classification dataset and benchmark［C］∥ IEEE/CVF Conference on Computer Vision and Pattern Recognition. Nashville, TN: IEEE，2021：13456-13467. ［12］ LIU W，ANGUELOV D，ERHAN D，et al. SSD：single shot MultiBox detector［M］∥Lecture Notes in Computer Science. Cham：Springer International Publishing，2016：21-37.

相似文献/References:

[1]陈炳煌.车前小型障碍物图像检测与分类方法[J].福建理工大学学报,2020,18(01):57.[doi:10.3969/j.issn.1672-4348.2020.01.011]
　CHEN Binghuang.Image detection and classification of small obstacles in front of vehicles[J].Journal of Fujian University of Technology;,2020,18(06):57.[doi:10.3969/j.issn.1672-4348.2020.01.011]
[2]喻露,戴甜杰,余丽华.基于改进YOLOv5的道路病害智能检测[J].福建理工大学学报,2023,21(04):332.[doi:10.3969/j.issn.1672-4348.2023.04.005]
　YU Lu,DAI Tianjie,YU Lihua.Automatic detection of pavement defect based on improved YOLOv5 algorithm[J].Journal of Fujian University of Technology;,2023,21(06):332.[doi:10.3969/j.issn.1672-4348.2023.04.005]

更新日期/Last Update: 2024-12-25