臺灣博碩士論文加值系統

The detection of foreign objects on railway scenes is a critical aspect of autonomous train systems. Navigating and making informed decisions in real-time is important for autonomous systems. Identifying and promptly reacting to foreign objects that intrude onto railway tracks is crucial for maintaining the integrity of the train's operations and ensuring the safety of passengers, crew, and infrastructure. This research specifically focuses on the challenge of identifying foreign objects that intrude onto railway tracks. Due to the unpredictable nature of the objects that can appear on the tracks, traditional supervised learning methods are not applicable. This is because there is a lack of labeled data, and failure to detect uncommon or edge cases can be costly and must be avoided at all costs.

To address this problem, the Fully Convolutional Data Description (FCDD) technique is employed for anomaly detection. FCDD allows the model to classify anomalies effectively by learning the underlying data distribution without explicit anomaly labels. Additionally, the model's ability to explain anomalies is enhanced by utilizing heatmaps, which provide visual representations of the image regions that contribute to anomaly detection.

Considering the challenges in acquiring test data for this specific problem, the model's evaluation is performed on a dataset that combines real-world examples with synthetic images generated using a technique called CutPaste augmentation. This approach helps augment the dataset and diversify the anomaly instances for more comprehensive testing and evaluation. By leveraging FCDD and incorporating heatmap explanations, this work aims to tackle the task of detecting foreign objects on railway tracks in an effective and interpretable manner. The resulting model is a highly skilled classifier network capable of detecting and classifying any object as an anomaly.

The detection of foreign objects on railway scenes is a critical aspect of autonomous train systems. Navigating and making informed decisions in real-time is important for autonomous systems. Identifying and promptly reacting to foreign objects that intrude onto railway tracks is crucial for maintaining the integrity of the train's operations and ensuring the safety of passengers, crew, and infrastructure. This research specifically focuses on the challenge of identifying foreign objects that intrude onto railway tracks. Due to the unpredictable nature of the objects that can appear on the tracks, traditional supervised learning methods are not applicable. This is because there is a lack of labeled data, and failure to detect uncommon or edge cases can be costly and must be avoided at all costs.

To address this problem, the Fully Convolutional Data Description (FCDD) technique is employed for anomaly detection. FCDD allows the model to classify anomalies effectively by learning the underlying data distribution without explicit anomaly labels. Additionally, the model's ability to explain anomalies is enhanced by utilizing heatmaps, which provide visual representations of the image regions that contribute to anomaly detection.

Considering the challenges in acquiring test data for this specific problem, the model's evaluation is performed on a dataset that combines real-world examples with synthetic images generated using a technique called CutPaste augmentation. This approach helps augment the dataset and diversify the anomaly instances for more comprehensive testing and evaluation. By leveraging FCDD and incorporating heatmap explanations, this work aims to tackle the task of detecting foreign objects on railway tracks in an effective and interpretable manner. The resulting model is a highly skilled classifier network capable of detecting and classifying any object as an anomaly.

ABSTRACT i
Acknowledgments iii
Table of Contents iv
List of Tables vi
List of Figures vii
Chapter 1 Introduction 1
1.1 Background 1
1.2 Related Work 2
1.3 Research Objective 3
1.4 Limitations 4
Chapter 2 Literature Review 5
2.1 Machine Vision 5
2.2 Automated Defect Detection 5
2.3 Unsupervised Learning 8
2.4 Anomaly Detection 9
2.5 Object Detection and Segmentation 11
2.6 Semi-Supervised Learning 14
2.7 Cut Paste Augmentation 15
Chapter 3 Methodology 17
3.1 Model Architecture 17
3.2 Model Working 19
3.3 Dataset 20
3.4 Model Training 22
Chapter 4 Results 23
4.1 Experimental Setup 23
4.2 Evaluation Metrics 23
4.3 Results 25
4.3.1 Unsupervised Method 26
4.3.2 Semi-Supervised Method 29
Chapter 5 Conclusions and Future Work 35
5.1 Conclusions 35
5.2 Future Work 36
References 37

[1] P. Liznerski, L. Ruff, R. A. Vandermeulen, B. J. Franks, M. Kloft, and K.-R. Müller, “Explainable deep one-class classification,” 2020, arXiv:2007.01760. [Online]. Available: http://arxiv.org/abs/1906.12340, Accessed: Mar. 18. 2021.

[2] L. Ruff, R. Vandermeulen, N. Goernitz, L. Deecke, S.A. Siddiqui, A. Binder, E. Müller and M. Kloft, "Deep one-class classification." in Proc. of Int. Conf. on Machine Learning, Stockholm, Sweden,pp.80:4393-4402, July, 2018.

[3] L. Ruff, R.A. Vandermeulen, N. Görnitz, A. Binder, E. Müller, K.R. Müller, and M. Kloft, "Deep semi-supervised anomaly detection," 2019, arXiv:1906.02694. [Online]. Available: http://arxiv.org/abs/1906.02694, Accessed: Feb. 14. 2020.

[4] O. Zendel, M. Murschitz, M. Zeilinger, D. Steininger, S. Abbasi, and C. Beleznai., "RailSem19: A Dataset for Semantic Rail Scene Understanding," in Proc. of IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), CA,
USA,pp.1221-1229, June, 2019.

[5] A. Hornberg, Handbook of Machine Vision. Wiley-VCH, 2006.

[6] V. Chandola, A. Banerjee, and V. Kumar, “Anomaly detection: A survey,” ACM
Computing Surveys, vol. 41, no. 3, pp.1-58, Jul. 2009.

[7] O. Rippel and D. Merhof, “Anomaly detection for automated visual inspection: a review,” Bildverarbeitung in der Automation,pp.1–13, March, 2023.

[8] D. Bank, N. Koenigstein, and R. Giryes, “Autoencoders,” 2021, arXiv:2003.05991.
[Online]. Available: https://arxiv.org/abs/2003.05991, Accessed: Apr. 03, 2021.

[9] S. S. A. Zaidi, M. S. Ansari, A. Aslam, N. Kanwal, M. Asghar, and B. Lee, “A Survey of Modern Deep Learning based Object Detection Models,” 2021, arXiv:2104.11892. [Online]. Available: https://arxiv.org/abs/2104.11892, Accessed: May 12, 2021.

[10] S. Minaee, Y. Boykov, F. Porikli, A. Plaza, N. Kehtarnavaz, and D. Terzopoulos, “Image Segmentation Using Deep Learning: A Survey,” 2020, arXiv:2001.05566. [Online].
Available: https://arxiv.org/abs/2001.05566, Accessed: Nov. 14, 2020.

[11] Y. Ouali, C. Hudelot, and M. Tami, “An Overview of Deep Semi-Supervised Learning,” 2020, arXiv:2006.05278. [Online]. Available: https://arxiv.org/abs/2006.05278, Accessed: Jul. 06, 2020.

[12] D. Hendrycks, M. Mazeika, S. Kadavath, and D. Song, “Using self-supervised learning can improve model robustness and uncertainty,” 2019, arXiv:1906.12340. [Online]. Available: http://arxiv.org/abs/1906.12340, Accessed: Oct. 29, 2019.

[13] C.-L. Li, K. Sohn, J. Yoon, and T. Pfister, "CutPaste: Self-Supervised Learning for Anomaly Detection and Localization," 2021, arXiv:2104.04015. [Online]. Available:
https://arxiv.org/abs/2104.04015, Accessed: Apr. 08, 2021.

[14] G. Ghiasi, Y. Cui, A. Srinivas, R. Qian, T.Y. Lin, E.D. Cubuk, Q.V. Le, and B. Zoph., “Simple copy-paste is a strong data augmentation method for instance segmentation,” 2020, arXiv:2012.07177. [Online]. Available: http://arxiv.org/abs/2012.07177, Accessed: Jun. 23, 2021.

[15] J. Long, E. Shelhamer, and T. Darrell, “Fully convolutional networks for semantic segmentation,” 2015, arXiv:1411.4038. [Online]. Available: https://arxiv.org/abs/1411.4038, Accessed: Mar. 08, 2015.

[16] A. Kirillov, E. Mintun, N. Ravi, H. Mao, C. Rolland, L. Gustafson, T. Xiao, S. Whitehead, A.C. Berg, W.Y. Lo, and P. Dollár., “Segment anything,” 2023, arXiv.2304.02643. [Online]. Available: http://arxiv.org/abs/2304.02643, Accessed: Apr. 5, 2023.

[17] Ž. Ð. Vujovic, “Classification Model Evaluation Metrics,” IJACSA, vol. 12, no. 6, pp.599-606, 2021, doi: 10.14569/IJACSA.2021.0120670.