改进YOLOv7-Tiny的道路裂缝检测算法

doi:10.3778/j.issn.1002-8331.2404-0256

摘要/Abstract

摘要： 道路裂缝检测是道路工程中的重要环节。针对现阶段道路裂缝检测算法中准确度低、效率低的问题，提出了一种基于YOLOv7-Tiny的轻量型道路裂缝检测算法YOLOv7-TPSF。引入部分卷积PConv，对原网络中耗参量较多的3×3卷积层进行部分替换，降低模型的参数量，提升模型的训练速度；结合特征融合网络BiFusion Neck与加权特征金字塔BiFPN的优点，提出了新的特征融合模块Bi-FusFPN，减少网络计算量，强化多尺度特征的融合能力；在输出端添加无参注意力机制SimAM，进一步提高大、中、小三类目标的检测能力。实验结果表明，YOLOv7-TPSF算法相较于YOLOv7-Tiny算法，网络参数量与计算量分别减少了31.7%、34.6%，准确度与检测速度分别提高了3.7%、9.7%，一定程度上满足了道路裂缝检测准确性与实时性的需求。

关键词: 道路裂缝检测, YOLOv7-Tiny, 轻量型, 注意力机制, 特征融合模块Bi-FusFPN

Abstract: In terms of road engineering, road crack detection plays an important role. Aiming at the problems of low accuracy and efficiency in current road crack detection algorithms, a road crack detection lightweight algorithm YOLOv7-TPSF based on YOLOv7-Tiny is proposed. Partial convolution PConv is used to replace some 3×3 convolution layers with more consumption parameters in the original network to reduce the number of model parameters and improve the training speed of the model. Combining the advantages of the feature fusion network BiFusion Neck and the weighted bidirectional feature pyramid BiFPN, a new feature fusion module Bi-FusFPN is proposed to reduce network computation and strengthen the ability of multi-scale feature fusion. A non-parametric attention mechanism SimAM is added at the output end to further improve the detection ability of large, medium, and small targets. The experimental results show that compared to the YOLOv7-Tiny algorithm, the number of network parameters and computations in the YOLOv7-TPSF algorithm are reduced by 31.7% and 34.6%, the accuracy and detection speed are improved by 3.7% and 9.7%, meeting the requirements for accuracy and real-time detection of road cracks to a certain extent.

Key words: road crack detection, YOLOv7-Tiny, lightweight, attention mechanism, feature fusion module Bi-FusFPN

王启涵, 刘超. 改进YOLOv7-Tiny的道路裂缝检测算法[J]. 计算机工程与应用, 2025, 61(10): 372-380.

WANG Qihan, LIU Chao. Improved YOLOv7-Tiny Road Crack Detection Algorithm[J]. Computer Engineering and Applications, 2025, 61(10): 372-380.

参考文献

[1] REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: unified, real-time object detection[C]//Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2016: 779-788.
[2] WEI L, DRAGOMIR A, DUMITRU E, et al. SSD: single shot multibox detector[C]//Proceedings of the 14th European Conference on Computer Vision. Cham: Springer, 2016: 21-37.
[3] ZHANG J, QIAN S R, TAN C. Automated bridge crack detection method based on lightweight vision models[J]. Complex & Intelligent Systems, 2023, 9(2): 1639-1652.
[4] 闫卫坡, 王志斌, 任祖跃, 等. 基于改进YOLOv5模型的高速道路裂缝检测研究[J]. 软件导刊, 2023, 22(10): 191-197.
YAN W P, WANG Z B, REN Z Y, et al. Research on expressway crack detection based on improved YOLOv5 model[J]. Software Guide, 2023, 22(10): 191-197.
[5] YU G, ZHOU X L. An improved YOLOv5 crack detection method combined with a bottleneck transformer[J]. Mathematics, 2023, 11(10): 2377.
[6] 刘浩翰, 樊一鸣, 贺怀清, 等. 改进YOLOv7-tiny的目标检测轻量化模型[J]. 计算机工程与应用, 2023, 59(14): 166-175.
LIU H H, FAN Y M, HE H Q, et al. Improved YOLOv7-tiny’s object detection lightweight model[J]. Computer Engineering and Applications, 2023, 59(14): 166-175.
[7] 王博, 李齐, 刘皎. 一种轻量级的SSD道路裂缝检测算法[J]. 商洛学院学报, 2022, 36(4): 83-90.
WANG B, LI Q, LIU J. A lightweight SSD road crack detection algorithm[J]. Journal of Shangluo University, 2022, 36(4): 83-90.
[8] GIRSHICK R. Fast R-CNN[C]//Proceedings of the 2015 IEEE International Conference on Computer Vision. Piscataway: IEEE, 2015: 1440-1448.
[9] HE K M, GKIOXARI G, DOLLáR P, et al. Mask R-CNN[C]//Proceedings of the 2017 IEEE International Conference on Computer Vision. Piscataway: IEEE, 2017: 2980-2988.
[10] 王欣杰. 基于改进U-Net的道路裂缝图像分割方法研究[D]. 杭州: 杭州电子科技大学, 2023.
WANG X J. Research on road crack image segmentation method based on improved U-Net[D]. Hangzhou: Hangzhou Dianzi University, 2023.
[11] KANG D, BENIPAL S S, GOPAL D L, et al. Hybrid pixel-level concrete crack segmentation and quantification across complex backgrounds using deep learning[J]. Automation in Construction, 2020, 118: 103291.
[12] YANG L X, ZHANG R Y, LI L D, et al. SimAM: a simple, parameter-free attention module for convolutional neural networks[C]//Proceedings of the 3rd International Conference on Machine Learning, 2021: 11863-11874.
[13] WANG C Y, BOCHKOVSKIY A, LIAO H M. YOLOv7: trainable bag-of-freebies sets new state-of-the-art for real-time object detectors[C]//Proceedings of the 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2023: 7464-7475.
[14] JOCHER G, CHAURASIA A, STOKEN A, et al. YOLOv5: v6.2-YOLOv5 classification models, Apple M1, reproducibility, ClearML and Deci. ai integrations[EB/OL]. （2022-08-17）[2024-01-15]. https://zenodo.org/record/7002879.
[15] LIN T Y, DOLLáR P, GIRSHICK R, et al. Feature pyramid networks for object detection[C]//Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2017: 936-944.
[16] LIU S, QI L, QIN H F, et al. Path aggregation network for instance segmentation[C]//Proceedings of the 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2018: 8759-8768.
[17] CHEN J R, KAO S H, HE H, et al. Run, don’t walk: chasing higher FLOPS for faster neural networks[C]//Proceedings of the 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2023: 12021-12031.
[18] LI C Y, LI L L, GENG Y F, et al. YOLOv6 v3. 0: a full-scale reloading[J]. arXiv:2301.05586, 2023.
[19] TAN M X, PANG R M, LE Q V. EfficientDet: scalable and efficient object detection[C]//Proceedings of the 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2020: 10778-10787.
[20] ARYA D, MAEDA H, GHOSH S K, et al. Crowdsensing-based road damage detection challenge (CRDDC’2022)[C]//Proceedings of the 2022 IEEE International Conference on Big Data. Piscataway: IEEE, 2022: 6378-6386.
[21] REDMON J, FARHADI A. YOLOv3: an incremental improvement[J]. arXiv:1804.02767, 2018.
[22] BOCHKOVSKIY A, WANG C Y, LIAO H M. YOLOv4: optimal speed and accuracy of object detection[J]. arXiv:2004.10934, 2020.