THE ALGORITHM FOR DETECTING OIL STAINS AND FOREIGN OBJECTS ON THE BOTTOM OF EMU CARS BASED ON DEEP LEARNING
Keywords:
Multiple-unit train, Oil contamination detection, YOLOv12, DeepLabV3+, Data augmentation, Semantic segmentationAbstract
The operation safety of EMU (Electric Multiple Unit) trains is the core guarantee of the high-speed railway transportation system. Currently, the widely deployed EMU operation fault dynamic image detection system (TEDS) in China mainly relies on manual image interpretation, resulting in low detection efficiency, prone to missed detections and false detections. To address the challenges in detecting oil stains under EMU car bottoms, such as scarce samples, insufficient accuracy of a single model, and difficulty in identifying reflective oil stains, this paper proposes an oil stain detection algorithm based on the YOLOv12 and DeepLabV3+ dual-model collaboration. In terms of data augmentation, to address the deficiency of only 356 original oil stain samples, this paper designs a three-stage data augmentation strategy. This strategy expands the training set to 6786 images through basic geometric transformations, noise addition and blurring processing, as well as a composite enhancement pipeline based on the Albumentations library, effectively enhancing the generalization ability of the model. This paper uses the YOLOv12 as the target detection model and trains an oil stain detector on the expanded dataset. Experimental results show that the YOLOv12 model achieves an accuracy of0.88 on the validation set, a recall rate of 0.70, and a recall rate of 0.79 for the oil stain category, effectively identifying most oil stain targets. The oil stain candidate regions detected by YOLOv12 are input into the DeepLabV3+ network, using MobileNetV2 as a lightweight backbone network, and training a pixel-level oil stain segmentation model. Experimental results show that the model achieves an average IoU of 0.4535 and an average Dice coefficient of 0.4872 on the validation set. The test results show that the model can effectively identify reflective oil stains that are difficult for humans to distinguish and reduces false alarms for dried traces. The joint detection framework combining target detection and semantic segmentation proposed in this paper integrates the characteristics of YOLOv12’s rapid localization and DeepLabV3+’s fine segmentation. This model can adapt to the oil stain defect detection requirements in complex EMU operation environments and provide a reference path for research on railway image intelligent detection technologies.References
[1] Fan L, Li D Y, Liu B, et al. Research on core technology of TEDS image aided recognition for multiple units based on improved SOLOv2 network. Railway Locomotive & Car, 2025, 45(2): 1-11.
[2] Luo H. Research on intelligent recognition technology of fault image of high-speed EMU components. Beijing: Beijing Jiaotong University, 2022.
[3] Yang Z H. Research on catenary equipment defect identification based on deep learning. Beijing: China Academy of Railway Sciences, 2022.
[4] He K, Zhang X, Ren S, et al. Deep residual learning for image recognition//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas: IEEE, 2016: 770-778.
[5] Yin T Q, Zhou X H, Song L S. Extracting water body information from high-resolution remote sensing images by DeepLab V3+. Zhejiang Hydrotechnics, 2024, 52(6): 88-93.
[6] Zhu Y J, Cai G Q, Han J, et al. Small object detection in complex open-pit mine backgrounds based on improved YOLOv11. Industry and Mine Automation, 2025, 51(4): 93-99.
[7] Transportation Technology Center, Inc. Machine vision for railway inspection. Pueblo: TTCI, 2021.
[8] Falamarzi A, Moridpour S, Nazem M. A review on rail defect detection techniques. Australian Journal of Structural Engineering, 2019, 20(3): 201-213.
[9] Ren S, He K, Girshick R, et al. Faster R-CNN: Towards real-time object detection with region proposal networks. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(6): 1137-1149.
[10] Bochkovskiy A, Wang C Y, Liao H Y M. YOLOv4: Optimal speed and accuracy of object detection. arXiv preprint, arXiv:2004.10934, 2020.
[11] Wang A, Chen H, Liu L, et al. YOLOv10: Real-time end-to-end object detection. arXiv preprint, arXiv:2405.14458, 2024.
[12] Jocher G. YOLOv11. https://github.com/ultralytics/ultralytics, 2024.
[13] Tian Y, Ye Q, Doermann D. YOLOv12: Attention-centric real-time object detectors. arXiv preprint, arXiv:2502.12524, 2025.
[14] Long J, Shelhamer E, Darrell T. Fully convolutional networks for semantic segmentation//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Boston: IEEE, 2015: 3431-3440.
[15] Ronneberger O, Fischer P, Brox T. U-Net: Convolutional networks for biomedical image segmentation//International Conference on Medical Image Computing and Computer-Assisted Intervention. Munich: Springer, 2015: 234-241.
[16] Chen L C, Papandreou G, Kokkinos I, et al. DeepLab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected CRFs. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 40(4): 834-848.
[17] Chen L C, Zhu Y, Papandreou G, et al. Encoder-decoder with atrous separable convolution for semantic image segmentation//Proceedings of the European Conference on Computer Vision. Munich: Springer, 2018: 801-818
[18] Heckhel W, Helali A. Early detection and classification of Alzheimer’s disease through data fusion of MRI and DTI images using the YOLOv11 neural network. Frontiers in Neuroscience, 2025, 19: 1554015.
[19] LeCun Y, Bengio Y, Hinton G. Deep learning. Nature, 2015, 521(7553): 436-444.
[20] Pan S J, Yang Q. A survey on transfer learning. IEEE Transactions on Knowledge and Data Engineering, 2010, 22(10): 1345-1359.
[21] Redmon J, Divvala S, Girshick R, et al. You only look once: Unified, real-time object detection//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas: IEEE, 2016: 779- 788.
[22] Sandler M, Howard A, Zhu M, et al. MobileNetV2: Inverted residuals and linear bottlenecks//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Salt Lake City: IEEE, 2018: 4510-4520.