ERBW-bolt: A high-accuracy model for train underbody bolts detection in complex background

Abstract

Railways are developing rapidly in countries worldwide due to their efficient transportation capacity. Accurate detection of train underbody bolts is crucial to railway safety. However, current bolt detection methods have low accuracy and high computational complexity. At the same time, the complex bolt detection background affects the classification and positioning performance. To address these challenges, a computer vision detection algorithm ERBW-Bolt which can accurately detect bolts is proposed in this paper. Based on the YOLOv8 algorithm, first, the EMAC2f module with attention mechanism is introduced in the backbone network to enhance the feature extraction capability. In addition, the neck network is improved by Re-parameterized Bidirectional Feature Pyramid Network (RepBIFPN) to enhance the feature fusion capability and computational performance of the network. Finally, to improve the model’s ability to balance samples of different qualities, the bounding box loss function is replaced by Wise Intersection over Union (WIoU). We conducted various experiments on the train underbody bolt dataset. Experimental results show that compared with the YOLOv8, the mAP50 of ERBW-Bolt is improved by 2.6% to 85.8%, the mAP50:95 is improved by 2.5% to 55.9%, the FPS (Frames Per Second) is improved by 84.9 to 107.8 f/s, and the Parameters is reduced by 3.7M to 7.43M, which greatly improved detection accuracy and computational efficiency. Compared with Faster R-CNN (region-based convolutional neural network), SSD (single shot multibox detector), YOLOv5, YOLOv9, YOLOv10, and YOLOv11, the ERBW-Bolt achieves higher detection accuracy and computational efficiency. Compared with detection transformers (DETR) and real-time detection transformers (RTDETR), the ERBW-Bolt achieves a better balance between detection accuracy and computational efficiency. Overall, the ERBW-Bolt model satisfied the requirement for detecting train underbody bolts in complex backgrounds, which is of great significance to the detection and maintenance of train underbody bolts.

Keywords

Bolt detection attention mechanism feature pyramid network bounding box loss function

Get full access to this article

View all access options for this article.

References

Carion

Massa

Synnaeve

, et al. (2020) End-to-end object detection with transformers. In: 2020 European conference on computer vision (ECCV), Glasgow, 23–28 August, pp. 213–229. Cham: Springer.

Chen

Cheng

, et al. (2022a) A hybrid deep learning based framework for component defect detection of moving trains. IEEE Transactions on Intelligent Transportation Systems 23(4): 3268–3280.

Chen

Xiong

Yang

, et al. (2022b) Preload measurement of steel-to-timber bolted joint using piezoceramic-based electromechanical impedance method. Measurement 190: 110725.

Chu

Zhang

Wang

, et al. (2024) YOLO-O2E: A variant YOLO model for anomalous rail fastening detection. CMC: Computers Materials & Continua 80(1): 1143–1161.

Cui

n Y

, et al. (2023) Skip connection YOLO architecture for noise barrier defect detection using UAV-based images in high-speed railway. IEEE Transactions on Intelligent Transportation Systems 24(11): 12180–12195.

Ding

Zhang

, et al. (2021) RepVGG: Making VGG-style convnets great again. In: 2021 IEEE/CVF conference on computer vision and pattern recognition (CVPR), Nashville, TN, 19–25 June, pp. 13728–13737. New York: IEEE. Available at:

Dreisbach

Yokaribas

Dietrich

, et al. (2023) A novel data-based approach for monitoring fatigue crack propagation in welded joints under varying ambient temperature using the electromechanical admittance. Structural Health Monitoring: An International Journal 22(5): 3270–3285.

Gevorgyan

ZJA

(2022) SIoU loss: More powerful learning for bounding box regression. Available at: https://arxiv.org/abs/2205.12740

Girshick

Donahue

Darrell

, et al. (2014) Rich feature hierarchies for accurate object detection and semantic segmentation. In: 27th IEEE conference on computer vision and pattern recognition (CVPR), Columbus, OH, 23–28 June, pp. 580–587. New York: IEEE.

10.

Guo

Liu

, et al. (2022) Attention mechanisms in computer vision: A survey. Computational Visual Media 8(3): 331–368.

11.

Hou

Zhou

Feng

, et al. (2021) Coordinate attention for efficient mobile network design. In: 2021 IEEE/CVF conference on computer vision and pattern recognition (CVPR), Nashville, TN, 19–25 June, pp. 13713–13722. New York: IEEE.

12.

Hussain

(2024) YOLOv1 to v8: Unveiling each variant-a comprehensive review of YOLO. IEEE Access 12: 42816–42833.

13.

Khanam

Hussain

(2024) YOLOv11: An overview of the key architectural enhancements. Available at: https://arxiv.org/abs/2410.17725

14.

Kong

Zhu

SCM

, et al. (2018) Tapping and listening: A new approach to bolt looseness monitoring. Smart Materials and Structures 27(7): 07LT02.

15.

Gao

, et al. (2024) Bolt loosening angle measurement along full range of screw exposure length based on 3D point cloud. Automation in Construction 168: 105785.

16.

Liu

Qin

, et al. (2018) Path aggregation network for instance segmentation. In: 31st IEEE/CVF conference on computer vision and pattern recognition (CVPR), Salt Lake City, UT, 18–23 June, pp. 8759–8768. New York: IEEE.

17.

Liu

Anguelov

Erhan

, et al. (2016) SSD: Single shot multibox detector. In: 14th European conference on computer vision (ECCV), Amsterdam, 11–14 October, pp. 21–37. Cham: Springer.

18.

YJA

(2023) MPDIoU: A loss for efficient and accurate bounding box regression. Available at: https://arxiv.org/abs/2307.07662

19.

Meher

Mishra

Sunny

(2022) Impedance-based looseness detection of bolted joints using artificial neural network: An experimental study. Structural Control & Health Monitoring 29(10): e3049.

20.

Núñez

Hernández

FCL

Granados

JJR

(2024) Automatic surveillance of people and objects on railway tracks. International Journal of Interactive Multimedia and Artificial Intelligence. Epub ahead of print 4 August. DOI: 10.9781/ijimai.2024.08.004.

21.

Ouyang

Zhang

, et al. (2023) Efficient multi-scale attention module with cross-spatial learning. In: 2023 IEEE international conference on acoustics, speech and signal processing (ICASSP), Rhodes Island, 4–10 June, pp. 1–5. New York: IEEE.

22.

Zheng

, et al. (2022) Loosening bolts detection of bogie box in metro vehicles based on deep learning. IEICE Transactions on Information and Systems 105(11): 1990–1993.

23.

Redmon

Divvala

Girshick

, et al. (2016) You only look once: Unified, real-time object detection. In: 2016 IEEE conference on computer vision and pattern recognition (CVPR), Seattle, WA, 27–30 June, pp. 779–788. New York: IEEE.

24.

Ren

Girshick

, et al. (2017) Faster R-CNN: Towards real-time object detection with region proposal networks. IEEE Transactions on Pattern Analysis and Machine Intelligence 39(6): 1137–1149.

25.

Rezatofighi

Tsoi

Gwak

, et al. (2019) Generalized intersection over union: A metric and a loss for bounding box regression. In: 2019 IEEE/CVF conference on computer vision and pattern recognition (CVPR), Long Beach, CA, 16–20 June, pp. 658–666. New York: IEEE.

26.

Song

Wang

, et al. (2024) DL-YOLOX: Real-time object detection via adjustable dilated enhancement for autonomous driving scene. Transactions of the Institute of Measurement and Control 47: 1556–1569. Available at: https://www.doi.org/10.1177/01423312241239020

27.

Tan

Pang

, et al. (CVPR) EfficientDet: Scalable and efficient object detection. In: 2020 IEEE/CVF conference on computer vision and pattern recognition (CVPR), Seattle, WA, 14–19 June, pp. 10778–10787. New York: IEEE.

28.

Tong

Chen

, et al. (2023) Wise-IoU: bounding box regression loss with dynamic focusing mechanism. Available at: https://arxiv.org/abs/2301.10051

29.

Wang

Chen

Liu

, et al. (2024a) Yolov10: Real-time end-to-end object detection. Available at: https://arxiv.org/abs/2405.14458

30.

Wang

Zhou

Yan

, et al. (2023a) PENet-KD: Progressive enhancement network via knowledge distillation for rail surface defect detection. IEEE Transactions on Instrumentation and Measurement 72: 1–11.

31.

Wang

C-Y

Yeh

I-H

Liao

H-YM

(2024b) YOLOv9: Learning what you want to learn using programmable gradient information. Available at: https://arxiv.org/abs/2402.13616

32.

Wang

Chen

, et al. (2023b) Research on double-sided detection method of bolt looseness based on machine vision. Journal of the Brazilian Society of Mechanical Sciences and Engineering 45(2): 99.

33.

Yang

Zhang

, et al. (2023) Deformable convolution and coordinate attention for fast cattle detection. Computers and Electronics in Agriculture 211: 108006.

34.

Zhang

Lou

Wang

, et al. (2023) PA-DETR: End-to-end visually indistinguishable bolt defects detection method based on transmission line knowledge reasoning. IEEE Transactions on Instrumentation and Measurement 72: 1–14.

35.

Zhang

Yuen

(2022) Bolt damage identification based on orientation-aware center point estimation network. Structural Health Monitoring-an International Journal 21(2): 438–450.

36.

Zhang

Ren

Zhang

, et al. (2022) Focal and efficient IOU loss for accurate bounding box regression. Neurocomputing 506: 146–157.

37.

Zhao

, et al. (2023) DETRs beat YOLOs on real-time object detection. In: 2019 IEEE/CVF conference on computer vision and pattern recognition (CVPR), Seattle, WA, 17–21 June, pp. 16965–16974. New York: IEEE.

38.

Zheng

Wang

Liu

, et al. (2020) Distance-IoU Loss: Faster and better learning for bounding box regression. In: 34th AAAI conference on artificial intelligence (AAAI), New York, 7–12 February, pp. 12993–13000. Washington, DC: AAAI.

39.

Zhu

, et al. (2020) Deformable DETR: Deformable transformers for end-to-end object detection. Available at: https://arxiv.org/abs/2010.04159

40.

Zou

Chen

Shi

, et al. (2023) Object detection in 20 years: A survey. Proceedings of the IEEE 111(3): 257–276.

41.

Zuo

Dong

Yang

, et al. (2023) Energy harvesting solutions for railway transportation: A comprehensive review. Renewable Energy 202: 56–87.