油污染會對海洋生態環境產生物理及化學的嚴重影響，儘管通過各種技術、法規、預防措施等方式，在減少海域的洩溢油事故發生已取得重大進展，但事故一旦發生，仍然存在嚴重的油污擴散風險，導致生物及其生態環境遭受破壞。本研究假設海上油污事件的初期應變可由衛星或(無人)飛行載具以遙測方式蒐集事故現場附近海域的光學影像，因此擬運用人工智慧協助應變處理人員辨識影像中的油污分布狀況，以有效地掌握海上油污位置及擴散範圍，從而降低人員在海域現場探勘之風險。
本研究採用人工智慧深度卷積模型U-Net架構進行監督式學習的訓練，第一步利用不同超參數進行模型調整，並使用召回率(recall)、精確度(precision)、IoU (intersection over union)、F1-Score、及訓練時間進行辨識效能評估，最終選用ELU激活函數、AdaMax優化器、及cross entropy損失函數，搭配U-Net模型架構進行海上油污辨識的模型訓練，可得到最佳的測試結果。全部測試樣本的辨識正確率之平均值約九成，其中召回率可達92.2%、精確度93.9%、IoU 86.6%、及F1-Score 91.0%，代表有一些影像的少部分區域仍有錯誤識別。
本文將辨識結果之IoU小於90%的影像案例區分為誤判(false positive 或稱false alarm)及遺失(false negative或稱missing)二類影像，並進行辨識錯誤的原因探討。在誤判案例中，IoU皆為75%以上，判別錯誤最多的部分為油水混雜影像，其次是波紋影像；而在遺失案例中，IoU為40%~89%，判別錯誤最多的情況是油水混雜及厚薄油混雜影像，其次為具有波紋或油污反光之影像。
最後利用油污與海水在影像上的光學特性差異來檢視AI辨識結果的合理性，選取影像中的油污分別呈現三種不同顏色(黃色、黑色、與紅色油污)之案例，在影像內選定一條斷面，並將該斷面上各位置點所對應的RGB值畫出，對照U-Net模型辨識後的輸出結果，證實人工智慧的判別可以由三個光波段的強度變化來合理說明；然而若要以傳統影像辨識方法，並基於不同光波段在海水與油污的物理特徵變化，來辨識海上的油污與非油污區的作法並不容易，因此有必要借助人工智慧的協助。人工智慧的優勢是其深度學習模型辨識油污範圍的機制不單以一項RGB數值作判斷，而是在學習過程中會考慮多項相關變數且給與不同的權重，進而計算出不同像素的預測機率，才據以判斷是否為油污。
本研究成果證實人工智慧不僅可以取代傳統聚類方法辨識效能之不足，也能改善人力辨識時因經驗差異導致判別不同之問題，且可省去許多人力消耗與資源運用。未來海域油污事件發生後，如能利用(無人)飛行載具以航拍方式蒐集事故現場附近海域的光學影像，然後運用人工智慧協助應變處理人員判別油污在現場的分布狀況，不僅能有效地掌握海上油污位置及擴散範圍，更能降低人員風險，加速後續油污應變處理程序，將油污對環境的危害程度降至最低。

In this research, it is assumed that optical images of the sea near an accident site can be collected through remote measurements, where AI is proposed to help identify the scope of oil spills in specific images.
In this study, the U-Net deep learning architecture, evolved from the traditional convolution neural network, is adopted for segmentation of oil spill images. We select different hyperparameters for model adjustment and use recall, precision, the IoU, the F1-Score, and training time to evaluate the identification performance. After training and testing the AI model for marine oil spill identification, we found that the use of U-Net with the ELU activation function, the AdaMax optimizer, and the cross-entropy loss function led to the best results. The average recognition accuracy of all of the test samples was approximately 90%, among which the recall was as high as 92.2%; the precision was as high as 93.9%; the IoU was 86.6%, and the F1-Score was as high as 91.0%.
The differences in the optical characteristics of oil and sea water in optical images were also utilized in this study to examine and verify the rationality of the AI identification results. The established AI model not only can replace traditional clustering methods and improve the shortcomings of their insufficient identification performance, but can also effectively assist in confirming the location and spread of oil pollution. Therefore, it can speed up the follow-up oil spill emergency response process and minimize the amount of pollution entering the environment.

1. Ajadi, O. A., Meyer, F. J., Tello, M., & Ruello, G. (2018). Oil Spill Detection in Synthetic Aperture Radar Images Using Lipschitz-Regularity and Multiscale Techniques. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 11(7), 2389-2405.
2. Albawi, S., Mohammed, T.A., & Al-Zawi, S. (2017, August). Understanding of a convolutional neural network. In 2017 International Conference on Engineering and Technology (ICET) (pp. 1-6). IEEE.
3. Allen, J. & Walsh, B. (2008, May). Enhanced oil spill surveillance, detection and monitoring through the applied technology of unmanned air systems. In International oil spill conference (Vol. 2008, No. 1, pp. 113-120). American Petroleum Institute.
4. Alpers, W., Holt, B., & Zeng, K. (2017). Oil spill detection by imaging radars: Challenges and pitfalls. Remote Sensing of Environment, 201, 133-147.
5. Bell, S., Lawrence Zitnick, C., Bala, K., & Girshick, R. (2016). Inside-outside net: Detecting objects in context with skip pooling and recurrent neural networks. In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 2874-2883).
6. Borge, J. N. & Soares, C. G. (2000). Analysis of directional wave fields using X-band navigation radar. Coastal Engineering, 40(4), 375-391.
7. Brekke, C. & Solberg, A. H. (2005). Oil spill detection by satellite remote sensing. Remote sensing of environment, 95(1), 1-13.
8. Chen, T. & Lu, S. (2017). Subcategory-aware feature selection and SVM optimization for automatic aerial image-based oil spill inspection. IEEE Transactions on Geoscience and Remote Sensing, 55(9), 5264-5273.
9. Chiu, C. M., Huang, C. J., Wu, L. C., Zhang, Y. J., Chuang, L. Z. H., Fan, Y., & Yu, H. C. (2018). Forecasting of oil-spill trajectories by using SCHISM and X-band radar. Marine pollution bulletin, 137, 566-581.
10. Dozat, T. (2016). Incorporating nesterov momentum into adam.
11. Duchi, J., Hazan, E., & Singer, Y. (2011). Adaptive subgradient methods for online learning and stochastic optimization. Journal of machine learning research, 12(7).
12. El-Magd, I. A., Zakzouk, M., Abdulaziz, A. M., & Ali, E. M. (2020). The Potentiality of Operational Mapping of Oil Pollution in the Mediterranean Sea near the Entrance of the Suez Canal Using Sentinel-1 SAR Data. Remote Sensing, 12(8), 1352.
13. Farrington, J. W. (2013). Oil pollution in the marine environment i: inputs, big spills, small spills, and dribbles. Environment: Science and Policy for Sustainable Development, 55(6), 3-13.
14. Fingas, M. F. & Brown, C. E. (1997). Review of oil spill remote sensing. Spill Science & Technology Bulletin, 4(4), 199-208.
15. Fingas, M. F. & Brown, C. E. (2014). Review of oil spill remote sensing. Marine pollution bulletin, 83(1), 9-23.
16. Fushiki, T. (2011). Estimation of prediction error by using K-fold cross-validation. Statistics and Computing, 21(2), 137-146.
17. Genovez, P. C., Jones, C. E., Sant’Anna, S. J., & Freitas, C. C. (2019). Oil Slick Characterization Using a Statistical Region-Based Classifier Applied to UAVSAR Data. Journal of Marine Science and Engineering, 7(2), 36.
18. Gonzalez, R. C. & Woods, R. E. (2007). Image processing. Digital image processing, 2(1).
19. Goodman, R. (1994). Overview and future trends in oil spill remote sensing. Spill Science & Technology Bulletin, 1(1), 11-21.
20. Grüner, K., Reuter, R., & Smid, H. (1991). A new sensor system for airborne measurements of maritime pollution and of hydrographic parameters. GeoJournal, 24(1), 103-117.
21. Handbook, I. T. O. P. F. (2010). International Tanker Owners Pollution Federation.
22. Hazmat, N. O. A. A. (2012). Open water oil identification job aid for aerial observation. Office of Response and Restoration Job Aid.
23. Hochreiter, S. (1998). The vanishing gradient problem during learning recurrent neural nets and problem solutions. International Journal of Uncertainty, Fuzziness and Knowledge-Based Systems, 6(02), 107-116.
24. ITOPF (International Tanker Owners Pollution Federation). (2011). Effects of Oil Pollution on the Marine Environment.
25. Jiao, Z., Jia, G., & Cai, Y. (2019). A new approach to oil spill detection that combines deep learning with unmanned aerial vehicles. Computers & Industrial Engineering, 135, 1300-1311.
26 Kingma, D. P. & Ba, J. (2014). Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980.
27. Krestenitis, M., Orfanidis, G., Ioannidis, K., Avgerinakis, K., Vrochidis, S., & Kompatsiaris, I. (2019). Oil spill identification from satellite images using deep neural networks. Remote Sensing, 11(15), 1762.
28. Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2017). Imagenet classification with deep convolutional neural networks. Communications of the ACM, 60(6), 84-90.
29. Li, H., Fang, J., Liu, S., Liang, X., Yang, X., Mai, Z., Wang, T., Chen, Z., & Ni, D. (2019). CR-Unet: A Composite Network for Ovary and Follicle Segmentation in Ultrasound Images. IEEE Journal of Biomedical and Health Informatics, 24(4), 974-983.
30. Lin, T. Y., Maire, M., Belongie, S., Hays, J., Perona, P., Ramanan, D., Zitnick, C.L., Dollár, P., & Zitnick, C. L. (2014, September). Microsoft coco: Common objects in context. In European conference on computer vision (pp. 740-755). Springer, Cham.
31. Liu, P., Li, Y., Liu, B., Chen, P., & Xu, J. (2019). Semi-Automatic Oil Spill Detection on X-Band Marine Radar Images Using Texture Analysis, Machine Learning, and Adaptive Thresholding. Remote Sensing, 11(7), 756.
32. Mera, D., Cotos, J. M., Varela-Pet, J., & Garcia-Pineda, O. (2012). Adaptive thresholding algorithm based on SAR images and wind data to segment oil spills along the northwest coast of the Iberian Peninsula. Marine pollution bulletin, 64(10), 2090-2096.
33. Misra, D., (2019). Mish: A self regularized non-monotonic neural activation function. arXiv preprint arXiv:1908.08681.
34. Nieto-Hidalgo, M., Gallego, A. J., Gil, P., & Pertusa, A. (2018). Two-stage convolutional neural network for ship and spill detection using SLAR images. IEEE Transactions on Geoscience and Remote Sensing, 56(9), 5217-5230.
35. NOAA O.O.R.A.R., (2020, Feb. 24). Oil Types, in: NOAA (Ed.).
36. Osman A, B., Ovinis, M., Ismael, M. A., & Fakhruldin M, H. (2018, November). Color-based Oil Spill Image Segmentation. In AIP Conference Proceedings (Vol. 2035, p. 080004).
37. Peterson, C. H., Rice, S. D., Short, J. W., Esler, D., Bodkin, J. L., Ballachey, B. E., & Irons, D. B. (2003). Long-term ecosystem response to the Exxon Valdez oil spill. Science, 302(5653), 2082-2086.
38. Popular Science Staff (2014, July 11). Massive slick of anchovy shoal up for a day at the beach, Popular Science. Web site: https://www.popsci.com/blog-network/ladybits/massive-slick-anchovy-shoal-day-beach/
39. Prasad, G. & Anuprakash, M. V. V. S. (2016). Pollution due to oil spills in marine environment and control measures. IOSR J Environ Sci Toxicol Food Technol, 10, 1-8.
40. Qian, N. (1999). On the momentum term in gradient descent learning algorithms. Neural networks, 12(1), 145-151.
41. Radovic, M., Adarkwa, O., & Wang, Q. (2017). Object recognition in aerial images using convolutional neural networks. Journal of Imaging, 3(2), 21.
42. Ronneberger, O., Fischer, P., & Brox, T. (2015). U-net: Convolutional networks for biomedical image segmentation. In International Conference on Medical image computing and computer-assisted intervention (pp. 234-241). Springer, Cham.
43. Ronneberger, O., Fischer, P., & Brox, T. (2015, October). U-net: Convolutional networks for biomedical image segmentation. In International Conference on Medical image computing and computer-assisted intervention (pp. 234-241). Springer, Cham.
44. Ruder, S. (2016). An overview of gradient descent optimization algorithms. arXiv preprint arXiv:1609.04747.
45. Saadoun, I. M. (2015). Impact of oil spills on marine life. Emerging pollutants in the environment-current and further implications, 75-104.
46. Salisbury, J. W., D'Aria, D. M., & Sabins Jr, F. F. (1993). Thermal infrared remote sensing of crude oil slicks. Remote Sensing of Environment, 45(2), 225-231.
47. Samberg, A. (2005, May). Advanced oil pollution detection using an airborne hyperspectral lidar technology. In Laser Radar Technology and Applications X (Vol. 5791, pp. 308-317). International Society for Optics and Photonics.
48. Sharma, S. (2017). Activation functions in neural networks. Towards Data Science, 6.
49. Sutton, R., (1986). Two problems with back propagation and other steepest descent learning procedures for networks, Proceedings of the Eighth Annual Conference of the Cognitive Science Society, 1986, pp. 823-832.
50. Wang, W., Yang, Y., Wang, X., Wang, W., & Li, J. (2019). Development of convolutional neural network and its application in image classification: a survey. Optical Engineering, 58(4), 040901.
51. Xu, J., Cui, C., Feng, H., You, D., Wang, H., & Li, B. (2019). Marine Radar Oil-Spill Monitoring through Local Adaptive Thresholding. Environmental Forensics, 20(2), 196-209.
52. Zhang, B., Matchinski, E. J., Chen, B., Ye, X., Jing, L., & Lee, K. (2019). Marine oil spills—Oil pollution, sources and effects. In World seas: an environmental evaluation (pp. 391-406). Academic Press.
53. Zhang, F., Shao, Y., Tian, W., & Wang, S. (2008, July). Oil spill identification based on textural information of SAR image. In IGARSS 2008-2008 IEEE International Geoscience and Remote Sensing Symposium (Vol. 4, pp. IV-1308). IEEE.
54. Zhou, Z., Siddiquee, M. M. R., Tajbakhsh, N., & Liang, J. (2018). Unet++: A nested u-net architecture for medical image segmentation. In Deep Learning in Medical Image Analysis and Multimodal Learning for Clinical Decision Support (pp. 3-11). Springer, Cham.
55. Zhu, X., Li, Y., Feng, H., Liu, B., & Xu, J. (2015). Oil spill detection method using X-band marine radar imagery. Journal of Applied Remote Sensing, 9(1), 095985.
56. 楊家敦 (2019)。108年法國「海洋油污染及海運化學品污染應變人力養成訓練」。台灣中油股份有限公司。
57. 臺南市政府重大海洋油汙染緊急應變計畫 (2016) 台南市：臺南市政府
58. 管瑞平（2010）。東港籍漁船苗栗外海遇大浪破損起火，海巡救出5船員。中央通訊社。取自：https://www.cna.com.tw/news/firstnews/202011125007.aspx