由於噴泥現象是國內外常見發生於鐵路的危害之一，當鐵路路基受到含水量、火車載重等各種因素影響土壤狀態，使路基等結構穩定度與強度降低，進而影響鐵路使用，鐵路噴泥檢測方法愈趨成熟，隨著透地雷達檢測噴泥技術的提出，以地球物理方法對路基結構進行檢測，因此可以完整識別肉眼不可視區域及提前預防噴泥情況發生，然而透地雷達檢測後的數據，主要依靠人工判讀消耗大量人力成本且需要一定的專業與經驗，因此本研究提出基於深度學習的方法自動判讀噴泥分佈，將利用透地雷達檢測後的訊號反射圖作為辨識噴泥的影像資料進行深度學習訓練。
隨著深度學習發展快速，近年來計算機視覺在自動化檢測的應用相當廣泛，本研究應用深度學習技術於噴泥分佈的影像辨識，針對噴泥訊號反射影像提出以RepVGG改良U-Net模型及使用Mask R-CNN與其他語義分割模型進行深度學習訓練，模型訓練的表現證明此計算機視覺的分割任務可行性，有效改善傳統判讀噴泥分佈的準確性和便利性，本研究成果也顯示本研究提出的模型與其他經典圖像分割模型相比在預測的精準度有最佳的表現，透過模型預測結果提升鐵路檢修的效率，能快速且準確辨識噴泥範圍走向、發展趨勢及嚴重程度。

As mud pumping is one of the common hazards in railways both domestically and internationally. Various factors such as moisture content and train loading can affect the soil conditions of railway embankments, leading to a decrease in stability and strength. This affects the usability of railways. The detection methods for railway mud pumping have become more mature over time. With the introduction of ground-penetrating radar (GPR) technology for mud pumping detection, the detection of railway subgrade structures through geophysical methods has become possible. This enables the comprehensive identification of invisible areas and early prevention of mud pumping incidents. However, the data obtained from ground-penetrating radar requires labor-intensive manual interpretation and expertise. Therefore, this research proposes an automatic mud pumping detection method based on deep learning. It aims to automatically interpret the distribution of mud pumping by training deep learning models using the signal reflection images obtained from ground-penetrating radar as input data.
With the rapid development of deep learning, computer vision has been widely applied in automated detection. In this study, deep learning techniques were employed for the image recognition of mud pumping distribution. Specifically, an improved U-Net model based on RepVGG, as well as Mask R-CNN and other semantic segmentation models, were used for deep learning training using mud pumping signal reflection images. The performance of the trained models demonstrated the feasibility of computer vision for segmentation tasks and effectively improved the accuracy and convenience of traditional mud pumping distribution interpretation. The results of this study also showed that the proposed models outperformed other classical image segmentation models in terms of prediction accuracy. By leveraging the predicted results of the models, the efficiency of railway maintenance can be enhanced, allowing for the rapid and accurate identification of mud pumping coverage, development trends, and severity.

[1] Takatoshi, I. (1997). Measure for the stabilization of railway earth structure. Japan Railway Technical Service, 290.
[2] Duong, T. V., Tang, A. M., Cui, Y. J., Trinh, V. N., Dupla, J. C., Calon, N., ... & Robinet, A. (2013). Effects of fines and water contents on the mechanical behavior of interlayer soil in ancient railway sub-structure. Soils and foundations, 53(6), 868-878.
[3] Yong, W.A.N.G., Ling-Wei, K.O.N. G., & Yan-Li, W.A.N.G.(2014, December). Mechanism and control of subgrade mud pumping under the cyclic load of train. In 2014 International Conference on Mechanics and Civil Engineering (icmce-14) (pp. 369-374). Atlantis Press.
[4] Hugenschmidt, J. (2000). Railway track inspection using GPR. Journal of Applied Geophysics, 43(2-4), 147-155.
[5] Kuo, C., Hsu, C., Chen, Y., Wu, C., Wang, H., Chen, D., & Lin, Y. (2016). Using ground-penetrating radar to promote the investigating efficiency in mud pumping disaster of railways. Proc. Eng. Technol. Innov, 4, 49-51.
[6] Kuo, C. P., Hsu, C. H., Wu, C. W., Liu, P. L., & Chen, D. W. (2017). Study on the mechanism and inspection method of railway pumping. In Bearing Capacity of Roads, Railways and Airfields (pp. 2063-2068). CRC Press.
[7] Kuo, C. (2021). Ground-penetrating radar to investigate mud pumping distribution along a railway line. Construction and Building Materials, 290, 123104.
[8] C.H.Hsu.(2017) Using Characterized Grids Resolving Method for GPR on Investigating Railway Mud pumping, Minghsin University of Scienceand Technology. master’sthesis.
[9] LeCun, Y., Bottou, L., Bengio, Y., & Haffner, P. (1998). Gradient-based learning applied to document recognition. Proceedings of the IEEE, 86(11), 2278-2324.
[10] Long, J., Shelhamer, E., Darrell, T., & Berkeley, U. C. (2014). Fully convolutional networks for semantic segmentation. arXiv 2015. arXiv preprint arXiv:1411.4038.
[11] Rumelhart, D. E., Hinton, G. E., & Williams, R. J. (1986). Learning representations by back-propagating errors. nature, 323(6088), 533-536.
[12] Chen, T., Cai, Z., Zhao, X., Chen, C., Liang, X., Zou, T., & Wang, P. (2020). Pavement crack detection and recognition using the architecture of segNet. Journal of Industrial Information Integration, 18, 100144.
[13] Al-Huda, Z., Peng, B., Algburi, R. N. A., Al-antari, M. A., Rabea, A. J., & Zhai, D. (2023). A hybrid deep learning pavement crack semantic segmentation. Engineering Applications of Artificial Intelligence, 122, 106142.
[14] Lee, S. Y., Le, T. H. M., & Kim, Y. M. (2023). Prediction and detection of potholes in urban roads: Machine learning and deep learning based image segmentation approaches. Developments in the Built Environment, 13, 100109.
[15] Xu, Y., Li, D., Xie, Q., Wu, Q., & Wang, J. (2021). Automatic defect detection and segmentation of tunnel surface using modified Mask R-CNN. Measurement, 178, 109316.
[16] He, K., Gkioxari, G., Dollár, P., & Girshick, R. (2017). Mask r-cnn. In Proceedings of the IEEE international conference on computer vision (pp. 2961-2969).
[17] Ding, X., Zhang, X., Ma, N., Han, J., Ding, G., & Sun, J. (2021). Repvgg: Making vgg-style convnets great again. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition (pp. 13733-13742).
[18] Ronneberger, O., Fischer, P., & Brox, T. (2015). U-net: Convolutional networks for biomedical image segmentation. In Medical Image Computing and Computer-Assisted Intervention–MICCAI 2015: 18th International Conference, Munich, Germany, October 5-9, 2015, Proceedings, Part III 18 (pp. 234-241). Springer International Publishing.
[19] Weng, Y., Zhou, T., Li, Y., & Qiu, X. (2019). Nas-unet: Neural architecture search for medical image segmentation. IEEE access, 7, 44247-44257.
[20] Girshick, R. (2015). Fast r-cnn. In Proceedings of the IEEE international conference on computer vision (pp. 1440-1448).
[21] Li, Y., Qi, H., Dai, J., Ji, X., & Wei, Y. (2017). Fully convolutional instance-aware semantic segmentation. In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 2359-2367).
[22] Chen, L. C., Papandreou, G., Schroff, F., & Adam, H. (2017). Rethinking atrous convolution for semantic image segmentation. arXiv preprint arXiv:1706.05587.
[23] Zhao, H., Shi, J., Qi, X., Wang, X., & Jia, J. (2017). Pyramid scene parsing network. In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 2881-2890).
[24] He, K., Zhang, X., Ren, S., & Sun, J. (2016). Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 770-778).
[25] Lin, T. Y., Dollár, P., Girshick, R., He, K., Hariharan, B., & Belongie, S. (2017). Feature pyramid networks for object detection. In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 2117-2125).
[26] Wang, T. P., Chen, C. C., Tong, L. T., Chang, P. Y., Chen, Y. C., Dong, T. H., ... & Cheng, S. N. (2015). Applying FDEM, ERT and GPR at a site with soil contamination: A case study. Journal of Applied Geophysics, 121, 21-30.
[27] Tsai, C. C., & Lin, C. H. (2022). Review and Future Perspective of Geophysical Methods Applied in Nearshore Site Characterization. Journal of Marine Science and Engineering, 10(3), 344.
[28] Lin, C. H., Lin, C. P., Hung, Y. C., Chung, C. C., Wu, P. L., & Liu, H. C. (2018). Application of geophysical methods in a dam project: Life cycle perspective and Taiwan experience. Journal of Applied Geophysics, 158, 82-92.
[29] 林志平. (2012). 地球物理探測技術於水庫構造物安全檢查應用之研究.
[30] Wilk, S., & Li, D. (2022). A Deep investigation into the mechanisms and factors producing mud pumping. In Advances in Transportation Geotechnics IV: Proceedings of the 4th International Conference on Transportation Geotechnics Volume 2 (pp. 25-36). Springer International Publishing.
[31] Kuo, C., Shen, T., Liu, M., Yang, S., & Ji, Z. (2018, July). Estimating the influence of saturation on investigating fouled railway ballasts using ground-penetrating-radar. In IGARSS 2018-2018 IEEE International Geoscience and Remote Sensing Symposium (pp. 9359-9362). IEEE.