台灣的地理條件優勢造就島國周圍有數個國際航道，航運貿易也就此成為我國主要的經濟活動之一。然而，海上交通頻度與複雜度的增加，使得船舶碰撞或擱淺等事故的發生風險相對提高，進而容易衍生出海上洩溢油事件。海域油污染會因波浪、潮汐、海流、海水、風和陽光等作用，使油污染加速的擴大污染範圍，導致一發不可收拾。因此，海域洩溢油事故發生後，最重要的便是對於油污染的緊急應變措施，有效率且快速地掌握油污染位置與範圍，為油污染緊急應變中最重要也最關鍵的一項。傳統對於油污染之影像辨識演算法大都需要人工選定閥值，以區分油污與背景物的分界，為避免人工介入的不確定性，本研究擬以深度學習網路架構之影像分割技術，建立一適用於海域油污染偵測之AI模式，以協助緊急應變人員在海域洩溢油事件發生後，能快速且精準地掌握油污染擴散後的分布資訊。
AI模型的深度學習需要大量的情資，本研究首先建立海域油污染資料庫，影像來源為14個國際單位組織於網路上公開之海域油污染事件的相關圖資，以確保海域油污影像來源之公信力。資料庫中的海域光學影像經過品質篩選後，拆分成訓練集、驗證集、及測試集，再使用圖像擴充方法增加影像的數量與多樣化後，用以對所選用的四套網路模型(U-Net、UNet++、Attention-UNet、與Attention-UNet++)依序進行油污辨識之訓練、驗證、及測試。模型學習時分別選用各超參數(包含訓練時期、批次大小、學習率、優化器、激活函數)的數個合理值，相互搭配成兩百多組的超參數設定，並對各模型進行優化、評估、與比較，以掌握各模型經超參數調整後的最佳模式，在不同油污狀況與環境條件下的辨識效能，以及適用時機。最終選定之最優選海域油污染辨識模式為UNet++，其超參數組合為訓練時期300、批次大小8、自適應學習率1e-4~1e-5、優化器Adam、激活函數ReLU，油污辨識結果為平均準確度82.3%、平均精確度80%、平均召回率81.6%、F1-score80.8%、訓練時長10.8小時、推論時間為8ms/frame。
為進一步檢驗此一最優選模式的泛化能力，本研究建立五個測試集進行油污辨識效能檢驗，結果顯示最優選模式於五個資料集的油污辨識效能皆表現良好，尤其於測試集Test 2的困難影像中之準確度達85.9%之成效，說明此模式很適用於複雜影像中；但對測試集Taiwan的油污辨識精確度較低，因此未來將可在訓練集與驗證集中增加台灣近岸海域的油污染事故影像，強化模式在訓練階段的學習，以此提高其精確度。然而在測試集Taiwan的召回率卻高達77.1%，代表影像中是油污的條件下，有將近八成被正確地辨識出來，因此在國內發生洩溢油事件後，該最優選模式便可高效地協助緊急應變單位掌握油污的地點與範圍。本研究更依據油污染的邊界清晰度、油污的面積大小、及有結構物之影像等三種不同類型的案例，以四套模型分別經超參數優化後的各最佳化模式之影像分割結果進行比較，並探討各優化模式在不同場景下的油水分割效果之優劣。為提升人工標注的可信度，本研究亦利用光學影像的物理特性以雙重驗證人工標注的結果，進一步探討最優選模式的辨識結果。
本研究所建立之最優選海域油污染辨識模式經過嚴謹的深度學習過程與超參數調整、案例測試與比較後，證實不但可供相關緊急應變單位於油污染應變過程更有效率地掌握油污染位置與範圍，也能運用於長期作業的油污染例行性監測(或警報)系統中。

SUMMARY
The main purpose of this research is to establish an AI model which is essentially suitable for detecting oil spill specially in marine areas, in order to assist emergency response personnel in quickly and accurately grasping the distribution information of oil pollution after oil spill incidents.
The deep learning of the AI model basically requires a large amount of data. For this research, the marine oil spill database images were sourced from 14 reputable international units and organizations, which publicly share optical images of oil spill in marine environments on the internet. This certainly ensures that non-oil spill images are being avoided to prevent all sought of incorrect learning. The optical images of the marine areas in the database have undergone quality screening, allocation into datasets and augmentation to enhance both the quantity and diversity of the data. Four networks (U-Net, UNet++, Attention-UNet and Attention-UNet++) were selected. More than 200 sets of hyperparameters were sequentially combined, including (epoch, batch size, learning rate, optimizer and activation function) for training, validation and testing. After optimization, evaluation and comparison of each model, the chosen best model for marine oil spill detection is UNet++. Its hyperparameter combination is 300 epochs, 8 batch size, 1e-4~1e-5 adaptive learning rate, Adam optimizer and ReLU activation function. The average accuracy of oil spill detection is about 82.3%, average precision is of 80%, average recall is of 81.6%, and F1-score is around 80.8%. The training time is 10.8 hours and the inference time is 8ms/frame.
In this research, the RGB physical characteristics of images were utilized to verify the accuracy of manually labeling mask images and the recognition results of the best model. Through rigorous test, comparison and test on different test sets for generalization, it has been confirmed that the best model just not only allows relevant emergency response units to efficiently identify the location and extent of oil spill during response operations but also can be applied to long-term oil spill routine monitoring (or alarm) systems.

1. Annunciado, T. R., & Sydenstricker, T. H. D., Experimental investigation of various vegetable fibers as sorbent materials for oil spills, Marine Pollution Bulletin, 2005.
2. Alattas, R., Oil Spill Detection in SAR Images Using Minimum Cross-Entropy Thresholding, The 2014 7th International Congress on Image and Signal Processing, 2014.
3. Anderson, P., He, X., Buehler, C., Teney, D., Johnson, M., Gould, S., & Zhang, L.: Bottom-up and top-down attention for image captioning and vqa, arXiv preprint arXiv:1707.07998, 2017.
4. Albawi, S., & Al-zawi, S., Understanding of a Convolutional Neural Network, ICET2017, 2017.
5. Anandharajan, T. R. V., Jijendiran, R., Vignajeth, K. K., & Hariharan, G. A., Detection and Analysis of Oil Spill in Ocean for Reduced Complexity in Extraction Using Image Processing, Advances in Intelligent Systems and Computing, 2018.
6. Alvernia, P., Soesilo, T.E.B., & Herdiansyah, H., Social impacts of oil spills towards the people living in coastal areas, International Conference on Research Collaboration of Environmental Science, 2021.
7. Alzubaidi, L., Zhang, J., Humaidi, A.J., Al-Dujaili, A., Duan, Y., Al-Shamma, O., Santamaria, J., Fadhel, M. A., Al-Amidie, M., & Farhan, L., Review of deep learning: concepts, CNN architectures, challenges, applications, future directions, p. 53, J Big Data, 2021.
8. Burgherr, P., In-depth analysis of accidental oil spills from tankers in the context of global spill trends from all sources, J Hazard Mater, 140(1-2): p. 245-56, 2007.
9. Basit, A., Siddique, M. A., & Sarfraz, M. S., Deep Learning Based Oil Spill Classification Using Unet Convolutional Neural Network, in 2021 IEEE International Geoscience and Remote Sensing Symposium IGARSS, p. 3491-3494, 2021.
10. Buttar, K. P. & Sachan, M. K., Semantic segmentation of clouds in satellite images based on U-Net++ architecture and attention mechanism. Expert Systems with Applications, 2022.
11. Chen, T. & Lu S., Subcategory-Aware Feature Selection and SVM Optimization for Automatic Aerial Image-Based Oil Spill Inspection, IEEE Transactions on Geoscience and Remote Sensing, p.5264-5273, 2017.
12. Cantorna, D., Dafonte, C., Iglesias, A., & Arcay, B., Oil spill segmentation in SAR images using convolutional neural networks. A comparative analysis with clustering and logistic regression algorithms, Applied Soft Computing Journal 84 (2019) 105716, 2019.
13. Chen, J., Zhang, W., Huang, T., & Fei, Y., Oil spills from global tankers: Status review and future governance, Journal of Cleaner Production, 2019.
14. Cao, F. & Bao Q., A Survey On Image Semantic Segmentation Methods With Convolutional Neural Network, in 2020 International Conference on Communications, Information System and Computer Engineering (CISCE), p.458-462, 2020.
15. Conceição, M.R.A., Mendonça, L.F.F., Lentini, C.A.D., Cunha Lima, A.T., Lopes, J.M., Vasconcelos, R.N., Gouveia, M.B., & Porsani, M.J., SAR Oil Spill Detection System through Random Forest Classifiers, Remote Sensing, 2021.
16. Chen, P., Zhou, H., Li, Y., Liu, B., & Liu, P., Oil Spill Identification in Radar Images Using a Soft Attention Segmentation Model. Remote Sensing, 2022.
17. Chen, Y. & Wang, Z., Marine Oil Spill Detection from SAR Images Based on Attention U-Net Model Using Polarimetric and Wind Speed Information, Int J Environ Res Public Health, 2022.
18. Danuser, G., Computer vision in cell biology, Cell 147(5): p. 973-8, 2011.
19. Duchi, J. Hazan, E., & Singer, Y., Adaptive Subgradient Methods for Online Learning and Stochastic Optimization, Journal of Machine Learning Research 12 (2011), p.2121-2159, 2011.
20. Fan, J., Zhang, F., Zhao, D., & Wang, J., Oil Spill Monitoring Based on SAR Remote Sensing Imagery, Aquatic Procedia, p.112-118, 2015.
21. Garcia-Pineda, O., MacDonald, I., & Zimmer, B., Synthetic Aperture Radar Image Processing Using the Supervised Textural-Neural Network Classification Algorithm., IEEE, 2008.
22. Guo, H., Wei, G., & An, J., Dark Spot Detection in SAR Images of Oil Spill Using Segnet, Applied Sciences, 2018.
23. Ghorbani, Z., & Behzadan, A. H., Identification and Instance Segmentation of Oil Spills Using Deep Neural Networks, in Proceedings of the 5th World Congress on Civil, Structural, and Environmental Engineering, 2020.
24. Ghorbani, Z., & Behzadan, A. H., Monitoring offshore oil pollution using multi-class convolutional neural networks, p.117884, Environ Pollut, 2021.
25. Gajjar, P., Garg, M., Shah, V., Shah, P., & Das, A., Applicability analysis of attention U-Nets over vanilla variants for automated ship detection, Reports on Geodesy and Geoinformatics, p. 9-14, 2022.
26. HE, Y., Wen, X., Wang, J., Zhang, C., & Lan, Y., Fault recognition based on 3D UNet + + L3 convolutional neural network, Progress in Geophysics, 37(2):0607-0616, 2022.
27. Huanga, X., Zhang, B., Perrie, W., Lu, Y., & Wang, C., A novel deep learning method for marine oil spill detection from satellite synthetic aperture radar imagery, Marine Pollution Bulletin 179 (2022) 113666, 2022.
28. Huby, A.A., Sagban, R., and Alubady, R., Oil Spill Detection based on Machine Learning and Deep Learning: A Review, in 2022 5th International Conference on Engineering Technology and its Applications (IICETA), p. 85-90, 2022.
29. Huettel, M., Oil pollution of beaches, Current Opinion in Chemical Engineering, 2022.
30. Kingma, D. & Ba, J., Adam: A Method for Stochastic Optimization, Proceedings of the 3rd International Conference on Learning Representations (ICLR 2015), 2015.
31. Krizhevsky, A., Sutskever, I., & Hinton, G.E., ImageNet classification with deep convolutional neural networks, Communications of the ACM, p.84-90, 2017.
32. Krestenitis, M., Orfanidis, G., Ioannidis, K., Avgerinakis, K., Vrochidis, S., & Kompatsiaris, I., Oil Spill Identification from Satellite Images Using Deep Neural Networks, Remote Sensing, 2019.
33. Kerf, D. T., Gladines, J., Sels, S., & Vanlanduit, S., Oil Spill Detection Using Machine Learning and Infrared Images, Remote Sensing, 2020.
34. Li, H., Wang, C., Zhang, H., Wu, F., & Li, J., Oil Slick Spot Detection Using K-distribution Model of The Sea Background, IEEE, 2009.
35. Lin, K., Yang, H. F., Hsiao, J. H., & Chen, C. S., Deep Learning of Binary Hash Codes for Fast Image Retrieval, Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, p. 27-35, 2015.
36. Long, J., Berkeley, E.S.U., & Darrell, T. B., Fully Convolutional Networks for Semantic Segmentation, CVPR2015, 2015.
37. Lin, H., Shi, Z., & Zou, Z., Fully Convolutional Network with Task Partitioning for Inshore Ship Detection in Optical Remote Sensing Images, IEEE Geoscience and Remote Sensing Letters, VOL. 14, NO. 10, 2017
38. Li, Y., Yang, X., Ye, Y., Cui, L., Jia, B., Jiang, Z., & Wang, S., Detection of Oil Spill Through Fully Convolutional Network, in Geo-Spatial Knowledge and Intelligence, p. 353-362, 2018.
39. Li, C., Tan, Y., Chen, W., Luo, X., Gao, Y., Jia, X., & Wan, Z., Attention UNet++: A Nested Attention-Aware U-Net for Liver CT Image Segmentation, IEEE, 2020.
40. Mera, D., Fernández-Delgado, M., Cotos, J. M., Viqueira, J. R. R., & Barro, S., Comparison of a massive and diverse collection of ensembles and other classifiers for oil spill detection in SAR satellite images, Neural Computing and Applications, p.1101-1117, 2017.
41. Mira, D., Gil, P., Alacid, B., & Torres, F., Oil Spill Detection using Segmentation based Approaches, in Proceedings of the 6th International Conference on Pattern Recognition Applications and Methods, p. 442-447, 2017.
42. Miegebielle, V., Conche, B., Zhexuan, H., Peigen, X., & Killisly, C., Use of Remote Sensing Radar Images for Offshore Oil Slick Detection in Oil and Gas Domain: Manual and Automatic Intapretation, IEEE, 2019.
43. Musumeci, F., Rottondi, C., Nag, A., Macaluso, I., Zibar, D., Ruffini, M., & Tornatore, M., An Overview on Application of Machine Learning Techniques in Optical Networks, IEEE Communications Surveys & Tutorials, p.1383-1408, 2019.
44. Mahmoud, A. Mohamed, S., El-Khoriby, S. A., AbdelSalam, R. A., & El-Khodary, I. A., Oil Spill Identification based on Dual Attention UNet Model Using Synthetic Aperture Radar Images, Journal of the Indian Society of Remote Sensing, p. 121-133, 2022.
45. Nie, W., & Zhang, X., Detecting Marine Oil Spill Pollution Based on Borda count method of Ocean Water Surface image, IEEE, 2012.
46. Nasteski, V., An overview of the supervised machine learning methods, Horizons.B, p.51-62, 2017.
47. Oktay, O., Schlemper1, J., Folgoc1, L. L., Lee, M., Heinrich, M., Misawa, K., Mori, K., McDonagh, S., Hammerla, N. Y., Kainz, B., Glocker, B., & Rueckert, D., Attention U-Net: Learning Where to Look for the Pancreas, arXiv, 2018.
48. Patel, S., Deep Learning Models for Image Segmentation, IEEE, 2021.
49. Reed, M., Johansen, O., Brandvik, P. J., & Daling, P. S., Oil Spill Modeling towards the Close of the 20th Century: Overview of the State of the Art, Spill Science & Technology Bulletin, 1999.
50. Ronneberger, O., Fischer, P., & Brox, T., U-net: Convolutional networks for biomedical image segmentation. In International Conference on Medical image computing and computer-assisted intervention (pp. 234-241). Springer, Cham, 2015.
51. Rai, M. C. E., Aburaed, N., Saad, M., Ahmad, H., Mansoori, S., & Marshall, S., Integrating Deep Learning with Active Contour Models in Remote Sensing Image Segmentation, IEEE, 2020.
52. Rousso, R., Katz, N., Sharon, G., Glizerin, Y., Kosman, E., & Shuster, A., Automatic Recognition of Oil Spills Using Neural Networks and Classic Image Processing, Water, 2022.
53. Seitinger, P., & Baumgartner, A. J., Oil spill remediation: An integrated approach for in-situ site clean-up, SPE Health, Safety and Environment in Oil and Gas Exploration and Production Conference, 1994.
54. Singha, S., Vespe, M., & Trieschmann, O., Automatic Synthetic Aperture Radar based oil spill detection and performance estimation via a semi-automatic operational service benchmark, Mar Pollut Bull, p.199-209, 2013.
55. Sinha, Y., Jain, P., & Kasliwal, N., Comparative Study of Preprocessing and Classification Methods in Character Recognition of Natural Scene Images, Advances in Intelligent Systems and Computing 390, 2016.
56. Song, D., Zhen, Z., Wang, B., Li, X., Gao, L., Wang, N., Xie, T., & Zhang, T., A Novel Marine Oil Spillage Identification Scheme Based on Convolution Neural Network Feature Extraction From Fully Polarimetric SAR Imagery, IEEE, 2020.
57. Tello, M., Bonastre, R., Lopez-Martinez, C., Mallorqui, J., Danisi, A., Di Martino, G., Iodice, A., Ruello, G., & Riccio, D., Characterization of Local Regularity in SAR Imagery by means of Multiscale Techniques: Application to Oil Spill Detection, IEEE, 2007.
58. Tieleman, T. & Hinton, G. Lecture 6.5-rmsprop: Divide the Gradient by a Running Average of Its Recent Magnitude. COURSERA: Neural Networks for Machine Learning, p.26-31, 2012.
59. Tewari, S., & Sirvaiya, A., Oil Spill Remediation and its Regulation, International Journal of Research In Science & Engineering, 2015.
60. Tong, S., Liu, X., Chen, Q., Zhang, Z., & Xie, G., Multi-Feature Based Ocean Oil Spill Detection for Polarimetric SAR Data Using Random Forest and the Self-Similarity Parameter, Remote Sensing, 2019.
61. Wan, J. & Cheng, Y., Remote Sensing Monitoring of Gulf of Mexico Oil Spill Using ENVISAT ASAR Images, IEEE, 2014.
62. Wang, D., & Han, M., SA-U-Net++: SAR marine floating raft aquaculture identification based on semantic segmentation and ISAR augmentation, Journal of Applied Remote Sensing, 2021.
63. Xu, F.Y., An, X. Z. & Liu, W. Q., Oil Spill Detection in SAR Images based on Improved YOLOX-S, in 2022 International Conference on Computer Engineering and Artificial Intelligence (ICCEAI), p. 261-265, 2022.
64. Yekeen, S. T., & Balogun, A. L., I Advances in Remote Sensing Technology, Machine Learning and Deep Learning for Marine Oil Spill Detection, Prediction and Vulnerability Assessment, MDPI, 2020.
65. Zhou, Z., Siddiquee, M. M. D., Tajbakhsh, N., & Liang, J., UNet++: A Nested U-Net Architecture for Medical Image Segmentation, arXiv, 2018.
66. 莊士賢、周芷蘭、連震杰、徐新綸，U-Net架構應用於海上油污光學影像之人工智慧識別，第42屆海洋工程研討會論文集，第343-348頁，2020。
67. 莊士賢、蕭郁涵、連震杰，人工智慧之影像分割模型在海域油污光學影像的識別，第44屆海洋工程研討會論文集，第343-348頁，2022。
68. 董昌明、韓瑩、徐廣珺、張琪、謝文鴻、周書逸，人工智能海洋學基礎及應用，科學出版社，中國，第170-176頁，2022。
69. 重大海洋污染緊急應變計畫，海洋委員會，2022年。