STA-YOLOv7：基于Swin-Transformer改进YOLOv7算法用于道路异常病害检测
STA-YOLOv7: Swin-Transformer-Enabled YOLOv7 for Road Damage Detection
DOI: 10.12677/CSA.2023.135113, PDF,  被引量    科研立项经费支持
作者: 张冬梅, 徐志洁：北京建筑大学理学院，北京
关键词: 深度学习；YOLOv7；Swin-Transformer；AlignOTA；病害异常检测；Deep Learning； YOLOv7； Swin-Transformer； AlignOTA； Road Damage Detection
摘要: 本文提出了一种基于Swin-Transformer改进的YOLOv7道路异常病害检测方法(STA-YOLOv7)，旨在解决道路异常病害图像分辨率较高以及多尺度目标检测不准确等问题。具体地，该方法在YOLOv7的结构中嵌入了Swin-Transformer中基于滑动窗口的设计的编码器，以捕捉不同尺度下病害的上下文信息与全局依赖关系，充分学习目标的语义特征。此外，我们还引入了AlignOTA损失函数来为模型训练提供更精确的标签分配策略，增强分类与回归的一致性。通过与Swin-Transformer、YOLOv7、TPH-YOLOv5等算法进行比较，STA-YOLOv7能够有效检测不同目标，降低漏检率的同时提高了准确率，适用于不同环境下各种尺度病害的检测，达到了实际复杂未知场景中实时性应用的需求。
Abstract: We propose an improved YOLOv7 method for road damage detection based on Swin-Transformer (STA-YOLOv7), aiming to address the challenges of high-resolution road damage images and inaccurate multi-scale object detection. Specifically, the Swin-Transformer encoder based on a sliding window design is embedded into the YOLOv7 architecture to capture contextual information and global dependencies of damages at different scales, and fully learn the semantic features of the targets. In addition, the AlignOTA loss function is introduced to provide more precise label assignment strategies for model training, enhancing the consistency of classification and regression. Compared with Swin-Transformer, YOLOv7, TPH-YOLOv5 and other algorithms, STA-YOLOv7 can effectively detect different targets, reduce missed positives while improving accuracy, and is applicable for detecting anomalies of various scales in different environments. It meets the requirements of real-time application in complex and unknown scenarios.
文章引用：张冬梅, 徐志洁. STA-YOLOv7：基于Swin-Transformer改进YOLOv7算法用于道路异常病害检测[J]. 计算机科学与应用, 2023, 13(5): 1157-1165. https://doi.org/10.12677/CSA.2023.135113

参考文献

[1] Cristianinin, Taylorjs. 支持向量机导论[M]. 李国正, 王猛, 曾华军, 译. 北京: 电子工业出版社, 2004: 15-20.
[2] 楼竞. 基于图像分析的路面病害检测方法与系统开发[D]: [硕士学位论文]. 南京: 南京理工大学, 2008.
[3] 马梁, 苟于涛, 雷涛, 等. 基于多尺度特征融合的遥感图像小目标检测[J]. 光电工程, 2022, 49(4): 49-65.
[4] 刘宪明, 辛公锋. 国内基于深度学习的道路路面病害检测研究综述[J]. 电子世界, 2021(8): 96-98.
[5] 谷永立, 宗欣欣. 基于深度学习的目标检测研究综述[J]. 现代信息科技, 2022, 6(11): 76-81.
[6] Girshick, R., Donahue, J., Darrell, T. and Malik, J. (2014) Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation. 2014 IEEE Conference on Computer Vision and Pattern Recognition, Columbus, 23-28 June 2014, 580-587. [Google Scholar] [CrossRef] 
[7] Girshick, R.B. (2015) Fast R-CNN. 2015 IEEE International Confer-ence on Computer Vision (ICCV), Santiago, Chile, 7-13 December 2015, 1440-1448. [Google Scholar] [CrossRef] 
[8] Ren, S., He, K., Girshick, R. and Sun, J. (2016) Faster R-CNN: To-wards Real-Time Object Detection with Region Proposal Networks. IEEE Transactions on Pattern Analysis & Machine Intelligence, 39, 1137-1149. [Google Scholar] [CrossRef] 
[9] Redmon, J., Divvala, S., Girshick, R. and Farhadi, A. (2016) You Only Look Once: Unified, Real-Time Object Detection. 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, 27-30 June 2016, 779-788. [Google Scholar] [CrossRef] 
[10] Wang, C.Y., Bochkovskiy, A. and Liao, H.Y.M. (2022) YOLOv7: Trainable Bag-of-Freebies Sets New State-of-the-Art for Real-Time Object Detectors.
https://arxiv.org/abs/2207.02696
[11] Li, Z. and Zhou, F. (2017) FSSD: Feature Fusion Single Shot Multibox De-tector.
https://arxiv.org/abs/1712.00960
[12] Lin, T.-Y., Goyal, P., Girshick, R., He, K. and Dollár, P. (2017) Fo-cal Loss for Dense Object Detection. 2017 IEEE International Conference on Computer Vision (ICCV), Venice, 22-29 October 2017, 2999-3007. [Google Scholar] [CrossRef] 
[13] Vaswani, A., Shazeer, N., Parmar, N., et al. (2017) Attention Is All You Need. Proceedings of the 31st International Conference on Neural Information Processing Systems, Long Beach, 4-9 December 2017.
[14] Mehta, S. and Rastegari, M. (2021) MobileViT: Light-Weight, General-Purpose, and Mo-bile-Friendly Vision Transformer.
https://arxiv.org/abs/2110.02178
[15] Dosovitskiy, A., Beyer, L., Kolesnikov, A., et al. (2020) An Image Is Worth 16x16 Words: Transformers for Image Recognition at Scale.
https://arxiv.org/abs/2010.11929
[16] Touvron, H., Cord, M., Douze, M., et al. (2020) Training Data-Efficient Image Transformers & Distillation through Attention.
https://arxiv.org/abs/2012.12877
[17] Pan, X., Ge, C., Lu, R., et al. (2022) On the Integration of Self-Attention and Convolution. 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), New Orleans, 18-24 June 2022, 805-815. [Google Scholar] [CrossRef] 
[18] Ge, Z., Liu, S., Wang, F., Li, Z. and Sun, J. (2021) Yolox: Exceeding Yolo Series in 2021.
https://arxiv.org/abs/2107.08430
[19] Xu, X., Jiang, Y., Chen, W., et al. (2023) DAMO-YOLO: A Report on Real-Time Object Detection Design.
https://arxiv.org/abs/2211.15444
[20] 戚玲珑, 高建瓴. 基于改进YOLOv7的小目标检测[J]. 计算机工程, 2023, 49(1): 41-48.
[21] 赵元龙, 单玉刚, 袁杰. 改进YOLOv7与DeepSORT的佩戴口罩行人跟踪[J]. 计算机工程与应用, 2023, 59(6): 221-230.
[22] Liu, Z., Lin, Y., Cao, Y., et al. (2021) Swin Transformer: Hierarchical Vision Transformer using Shifted Windows. 2021 IEEE/CVF International Conference on Computer Vision (ICCV), Montreal, 10-17 October 2021, 9992-10002. [Google Scholar] [CrossRef] 
[23] Lu, S., Liu, X., He, Z., Karkee, M. and Zhang, X. (2022) Swin-Transformer-YOLOv5 for Real-Time Wine GrapeBunch Detection.
https://arxiv.org/abs/2208.14508
[24] Gong, H., Mu, T., Li, Q., et al. (2022) Swin-Transformer-Enabled YOLOv5 with Attention Mechanism for Small Object Detection on Satellite Images. Remote Sensing, 14, Article No. 2861. [Google Scholar] [CrossRef] 
[25] Zhu, X., Lyu, S., Wang, X. and Zhao, Q. (2021) TPH-YOLOv5: Im-proved YOLOv5 Based on Transformer Prediction Head for Object Detection on Drone-Captured Scenarios. 2021 IEEE/CVF International Conference on Computer Vision Workshops (ICCVW), Montreal, 11-17 October 2021, 2778-2788. [Google Scholar] [CrossRef] 
[26] 李姗. 基于YOLOv5的道路病害检测与分类研究[J]. 现代计算机, 2021, 27(35): 75-79.