裂流(rip current)常將泳客捲往外海，造成意外，裂流發生機制複雜，迄今仍難以預測其可能發生的時間與範圍，掌握裂流特性，更加瞭解裂流，有助於未來進行裂流預警。因此，本研究目的為從觀測影像中，辨識出裂流，作為裂流研究資料來源。近年利用人工智慧(AI)研究興盛，本研究係利用AI中的深度學習(Deep Learning, DL)技術辨識裂流，採用的方法以卷積神經網路(CNN)為模式基礎改良的更快速區域卷積神經網路(Faster Region-base Convolutional Neural Network, Faster R-CNN)。
本研究分析國內外裂流衛星影像共220張，另外也蒐集100張無裂流影像，共320張進行訓練與驗證。在Faster R-CNN模式建置的訓練過程中，引用了影像擴增(data augmentation, DA)方法增加訓練資料量，並完成超參數(hyperparameter)之率定，以建置裂流辨識模式。本研究探討模式建置過程中信心門檻值(confidence value)之決定，當設定為0.5時，準確性可達86%；另外，模式訓練過程中使用了影像擴增方法，可使辨識準確率提高14%。透過本研究Faster R-CNN模式的辨識結果與人工診斷結果比較驗證，顯示Faster R-CNN模式辨識準確率可達86%。再與前人研究結果比較，發現本研究研發模式的訓練影像較少即可達到與前人相近的準確性，證實了深度學習演算法具有正確辨識出影像中出現裂流的優異能力。

Rip currents are dangerous currents that result in many accidents by sweeping swimmers and beach goers. The mechanism of rip currents is too complicated to predict when and where rip currents might occur till the present moment. Knowing more characteristics and features about rip currents will be beneficial to their early warning in the future. Therefore, the aim of this study is to recognize rip currents on images and also be researching data. Research with artificial intelligence (AI) is a big hit around the world, thus, this paper uses deep learning (DL), a branch of AI, to identify rip currents. The DL method used is Faster Region-base Convolutional Neural Network (Faster R-CNN) which is an improved version based on another famous model, Convolutional Neural Network (CNN). In this study, the quantity of used images is 220 rip currents images and 100 of non-rip currents images from satellite. In the process of setting up Faster R-CNN, data augmentation (DA) was employed in order to increase the training data amount and finished the calibration of hyperparameters later. This study discuss how large confidence threshold is, as it is 0.5, accuracy would be 86%. Furthermore, using DA on images before training, the accuracy of detecting rip current could improve 14%, better than not using DA. With both the result of Faster R-CNN and identified them by eyes, the accuracy of recognition rip currents would achieve 86%. In addition, comparing with previous study that used the same model, although the research data amount in this paper is less than the previous one, both accuracy are about the same. It shows that this kind of DL algorithm has the excellent abilities to detect rip currents on images correctly.

[1] 林雪美(2007)，臺灣北海岸海灘安全性之地型動力學調查，2007年地理學者學術研討會，國立高雄師範大學，共7頁。
[2] 林雪美(2009)，宜蘭外澳海灘裂流判釋與遊憩安全告示牌設計，國立臺灣師範大學地理學系，交通部觀光局計畫。
[3] 郭平巧、許弘莒、張裕弦、劉景毅(2011)，臺南市海灘類型與離岸流分布之變動探討，第33屆海洋工程研討會論文集，第453-457頁。
[4] 蕭仁豪、溫志中、李宗霖、蕭坤欣(2014)，海水浴場海灘安全性評估研究，第36屆海洋工程研討會論文集，第141-146頁。
[5] 黃柏仁(2018)，一種基於Faster R-CNN的快速虹膜切割演算法，國立中央大學碩士論文。
[6] 陳奕瑄(2018)，以CNN進行植物圖片的辨識以及處理後的再辨識，國立臺灣大學碩士論文。
[7] 交通部中央氣象局(2021)，海岸裂流監測與預警技術研究(1/3)，研究計畫報告。
[8] Barmpoutis, P., Dimitropoulos, K., Kaza, K., & Grammalidis, N. (2019, May). Fire detection from images using faster R-CNN and multidimensional texture analysis. In ICASSP 2019-2019 IEEE Interna-
tional Conference on Acoustics, Speech and Signal Processing (ICASSP)(8301-8305). IEEE.
[9] Barron, J. L., Fleet, D. J., & Beauchemin, S. S. (1994). Performance of optical flow techniques. International journal of computer vision, 12(1), 43-77.
[10] Borge, J. N. (2013). Use of X-band marine radars as a remote sensing system to survey wind-generated waves. WIT Transactions on Ecology and the Environment, 169, 27-37.
[11] Bowen, A. J., Inman, D. L., & Simmons, V. P. (1968). Wave ‘set‐down’and set‐up. Journal of Geophysical Research, 73(8), 2569-2577.
[12] Brander, R. W. (1999). Field observations on the morphodynamic evolution of a low-energy rip current system. Marine geology, 157(3-4), 199-217.
[13] Brannstrom, C., Brown, H. L., Houser, C., Trimble, S., & Santos, A. (2015). “You can't see them from sitting here”: Evaluating beach user understanding of a rip current warning sign. Applied Geography, 56, 61-70.
[14] Brewster, B. C., Gould, R. E., & Brander, R. W. (2019). Estimations of rip current rescues and drowning in the United States. Natural Hazards and Earth System Sciences, 19(2), 389-397.
[15] Brighton, B., Sherker, S., Brander, R., Thompson, M., & Bradstreet, A. (2013). Rip current related drowning deaths and rescues in Australia 2004–2011. Natural hazards and earth system sciences, 13(4), 1069-1075.
[16] Bruneau, N., Bonneton, P., Castelle, B., & Pedreros, R. (2011). Modeling rip current circulations and vorticity in a high‐energy mesotidal‐macrotidal environment. Journal of Geophysical Research: Oceans, 116(C7).
[17] Buscombe, D., & Carini, R. J. (2019). A data-driven approach to classifying wave breaking in infrared imagery. Remote Sensing, 11(7), 859.
[18] Buscombe, D., & Ritchie, A. C. (2018). Landscape classification with deep neural networks. Geosciences, 8(7), 244.
[19] Bryan, K. R., Davies-Campbell, J., Hume, T. M., & Gallop, S. L. (2019). The influence of sand bar morphology on surfing amenity at New Zealand beach breaks. Journal of Coastal Research, 87(SI), 44-54.
[20] Byeon, Y. H., & Kwak, K. C. (2017, July). A performance comparison of pedestrian detection using faster RCNN and ACF. In 2017 6th IIAI International Congress on Advanced Applied Informatics (IIAI-AAI) (858-863). IEEE.
[21] Carpi, L., Mucerino, L., Bonello, G., Besio, G., & Ferrari, M. (2021). Rip currents investigation on a Ligurian pocket beach, NW Mediterranean.Estuarine, Coastal and Shelf Science, 262, 107579.
[22] Castelle, B., Almar, R., Dorel, M., Lefebvre, J. P., Sénéchal, N., Anthony, E. J., Laibi, R., Chuchla, R., & Penhoat, Y. D. (2014). Rip currents and circulation on a high-energy low-tide-terraced beach (Grand Popo, Benin, West Africa). Journal of Coastal Research, (70 (10070)), 633-638.
[23] Castelle, B., & Coco, G. (2012). The morphodynamics of rip channels on embayed beaches. Continental Shelf Research, 43, 10-23.
[24] Castelle, B., & Coco, G. (2013). Surf zone flushing on embayed beaches.Geophysical Research Letters, 40(10), 2206-2210.
[25] Castelle, B., Scott, T., Brander, R. W., & McCarroll, R. J. (2016). Rip current types, circulation and hazard. Earth-Science Reviews, 163, 1-21.
[26] Clark, D. B., Feddersen, F., & Guza, R. T. (2010). Cross‐shore surfzone tracer dispersion in an alongshore current. Journal of Geophysical Research: Oceans, 115(C10).
[27] Clark, D. B., Lenain, L., Feddersen, F., Boss, E., & Guza, R. T. (2014). Aerial imaging of fluorescent dye in the near shore. Journal of Atmospheric and Oceanic Technology, 31(6), 1410-1421.
[28] Cui, F., Ning, M., Shen, J., & Shu, X. (2022). Automatic recognition and tracking of highway layer-interface using Faster R-CNN. Journal of Applied Geophysics, 196, 104477.
[29] de Silva, A., Mori, I., Dusek, G., Davis, J., & Pang, A. (2021). Automated rip current detection with region based convolutional neural networks.Coastal Engineering, 166, 103859.
[30] Dalrymple, R. A., MacMahan, J. H., Reniers, A. J., & Nelko, V. (2011). Rip currents. Annual Review of Fluid Mechanics, 43, 551-581.
[31] Dhillon, A., & Verma, G. K. (2020). Convolutional neural network: a review of models, methodologies and applications to object detection. Progress in Artificial Intelligence, 9(2), 85-112.
[32] Engle, J., MacMahan, J., Thieke, R. J., Hanes, D. M., & Dean, R. G. (2002). Formulation of a rip current predictive index using rescue data. In Proc. National Conf. on Beach Preservation Technology, FSBPA.
[33] Girshick, R. (2015). Fast r-cnn. In Proceedings of the IEEE international Conference on Computer Vision (1440-1448).
[34] Girshick, R., Donahue, J., Darrell, T., & Malik, J. (2014). Rich feature hierarchies for accurate object detection and semantic segmentation. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (580-587).
[35] Halawa, L. J., Wibowo, A., & Ernawan, F. (2019, October). Face recognition using faster R-CNN with inception-V2 architecture for CCTV camera. In 2019 3rd International Conference on Informatics and Computational Sciences (ICICoS) (1-6). IEEE.
[36] Haller, M. C., Dalrymple, R. A., & Svendsen, I. A. (2002). Experimental study of nearshore dynamics on a barred beach with rip channels. Journal of Geophysical Research: Oceans, 107(C6), 14-1.
[37] Haller, M. C., Honegger, D., & Catalan, P. A. (2014). Rip current observations via marine radar. Journal of Waterway, Port, Coastal,and Ocean Engineering, 140(2), 115-124.
[38] Han, J., Zhang, D., Cheng, G., Liu, N., & Xu, D. (2018). Advanced deep-learning techniques for salient and category-specific object detection: a survey. IEEE Signal Processing Magazine, 35(1), 84-100.
[39] Holman, R. A., & Stanley, J. (2007). The history and technical capabilities of Argus. Coastal Engineering, 54(6-7), 477-491.
[40] Huntley, D. A., Hendry, M. D., Haines, J., & Greenidge, B. (1988). Waves and rip currents on a Caribbean pocket beach, Jamaica. Journal of Coastal Research, 69-79.
[41] Inch, K. (2014). Surf zone hydrodynamics: Measuring waves and currents. Geomorphological Techniques, 3, 1-13.
[42] Jiang, G. Q., Xu, J., & Wei, J. (2018). A deep learning algorithm of neural network for the parameterization of typhoon‐ocean feedback in typhoon forecast models. Geophysical Research Letters, 45(8), 3706-3716.
[43] Jiang, H., & Learned-Miller, E. (2017, May). Face detection with the faster R-CNN. In 2017 12th IEEE International Conference on Automatic Face & Gesture Recognition (FG 2017) (650-657). IEEE.
[44] Johnson, D., & Pattiaratchi, C. (2004). Transient rip currents and nearshore circulation on a swell‐dominated beach. Journal of Geophysical Research: Oceans, 109(C2).
[45] Kang, K., Ouyang, W., Li, H., & Wang, X. (2016). Object detection from video tubelets with convolutional neural networks. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (817-825).
[46] Kim, J. H., Batchuluun, G., & Park, K. R. (2018). Pedestrian detection based on faster R-CNN in nighttime by fusing deep convolutional features of successive images. Expert Systems with Applications, 114, 15-33.
[47] Klein, A.H., da, F., Santana, G. G., Diehl, F. L., & De Menezes, J. T. (2003). Analysis of hazards associated with sea bathing: results of five years work in oceanic beaches of Santa Catarina state, southern Brazil. Journal of Coastal Research, 107-116.
[48] Lascody, R. L. (1998). East central Florida rip current program. National Weather Digest, 22(2), 25-30.
[49] Leatherman, S. B., & Leatherman, S. P. (2017). Techniques for detecting and measuring rip currents. International Journal of Earth Science and Geophys, 3, 014.
[50] Lee, J., Kim, D. H., Lee, S., & Lee, J. L. (2016). Lagrangian observation of rip currents at haeundae beach using an optimal buoy type gps drifter. Journal of Coastal Research, 75 (10075), 1177-1181.
[51] Lima, E., Sun, X., Dong, J., Wang, H., Yang, Y., & Liu, L. (2017). Learning and transferring convolutional neural network knowledge to ocean front recognition. IEEE Geoscience and Remote Sensing Letters, 14(3), 354-358.
[52] Lin, T. Y., Maire, M., Belongie, S., Hays, J., Perona, P., Ramanan, D., Dollár, P & Zitnick, C. L. (2014, September). Microsoft coco: Common objects in context. In European Conference on Computer Vision (740-755). Springer, Cham.
[53] Lin, Z., Courbariaux, M., Memisevic, R., & Bengio, Y. (2015). Neural networks with few multiplications. arXiv preprint arXiv:1510.03009.
[54] Liu, B., Yang, B., Masoud-Ansari, S., Wang, H., & Gahegan, M. (2021). Coastal Image Classification and Pattern Recognition: Tairua Beach, New Zealand. Sensors, 21(21), 7352.
[55] Liu, Y., and Wu, C. H. (2019). Lifeguarding Operational Camera Kiosk System (LOCKS) for flash rip warning: Development and application.Coastal Engineering, 152, 103537.
[56] Liu, L., Ouyang, W., Wang, X., Fieguth, P., Chen, J., Liu, X., & Pietikäinen, M. (2018). Deep Learning for Generic Object Detection: A Survey. arXiv e-prints, arXiv-1809.
[57] Li, M., Zhang, T., Chen, Y., & Smola, A. J. (2014, August). Efficient mini-batch training for stochastic optimization. In Proceedings of the 20th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (661-670).
[58] Lushine, J. B. (1991). A study of rip current drownings and related weather factors. National Weather Digest, 16(3), 13-19.
[59] Ma, S., Huang, Y., Che, X., & Gu, R. (2020). Faster RCNN‐based detection of cervical spinal cord injury and disc degeneration. Journal of Applied Clinical Medical Physics, 21(9), 235-243.
[60] MacMahan, J. H., Thornton, E. B., Reniers, A. J., Stanton, T. P., & Symonds, G. (2008). Low-energy rip currents associated with small bathymetric variations. Marine Geology, 255(3-4), 156-164.
[61] MacMahan, J. H., Thornton, E. B., & Reniers, A. J. (2006). Rip current review. Coastal Engineering, 53(2-3), 191-208.
[62] Maity, M., Banerjee, S., & Chaudhuri, S. S. (2021, April). Faster r-cnn and yolo based vehicle detection: A survey. In 2021 5th International Conference on Computing Methodologies and Communication (ICCMC)(1442-1447). IEEE.
[63] Mansour, R. F., Escorcia-Gutierrez, J., Gamarra, M., Villanueva, J. A., & Leal, N. (2021). Intelligent video anomaly detection and classification using faster RCNN with deep reinforcement learning model. Image and Vision Computing, 112, 104229.
[64] Maryan, C. C. (2018). Detecting Rip Currents from Images. University of New Orleans.
[65] Maryan, C., Hoque, M. T., Michael, C., Ioup, E., & Abdelguerfi, M. (2019). Machine learning applications in detecting rip channels from images.Applied Soft Computing, 78, 84-93.
[66] Meadows, G. A., Grimm, A., Brooks, C. N., & Shuchman, R. A. (2015). Remote sensing-based detection and monitoring of dangerous nearshore currents.
[67] Mucerino, L., Carpi, L., Schiaffino, C. F., Pranzini, E., Sessa, E., & Ferrari, M. (2021). Rip current hazard assessment on a sandy beach in Liguria, NW Mediterranean. Natural Hazards, 105(1), 137-156.
[68] Nelko, V., & Dalrymple, R. A. (2011). ‘Rip current prediction in ocean city, Maryland. Rip currents: Beach safety, physical oceanography, and wave modeling, Florida: CRC Press International, 45-57.
[69] Papageorgiou, C. P., Oren, M., & Poggio, T. (1998, January). A general framework for object detection. In Sixth International Conference on Computer Vision (IEEE Cat. No. 98CH36271) (555-562). IEEE.
[70] Philip, S., & Pang, A. (2016, June). Detecting and Visualizing Rip Current Using Optical Flow. In EuroVis (Short Papers) (19-23).
[71] Pitman, S., Gallop, S. L., Haigh, I. D., Mahmoodi, S., Masselink, G., & Ranasinghe, R. (2016). Synthetic imagery for the automated detection of rip currents. Journal of Coastal Research, (75 (10075)), 912-916.
[72] Pritchard, D. W., & Carpenter, J. H. (1960). Measurements of turbulent diffusion in estuarine and inshore waters. Hydrological Sciences Journal,5(4), 37-50.
[73] Rashid, A. H., Razzak, I., Tanveer, M., & Robles-Kelly, A. (2020, November). Ripnet: A lightweight one-class deep neural network for the identification of rip currents. In International Conference on Neural Infor-mation Processing (172-179). Springer, Cham.
[74] Redmon, J., Divvala, S., Girshick, R., & Farhadi, A. (2016). You only look once: Unified, real-time object detection. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (779-788).
[75] Ren, S., He, K., Girshick, R.B., & Sun, J. (2015). Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks. IEEE Transactions on Pattern Analysis and Machine Intelligence, 39, 1137-1149.
[76] Ren, S., He, K., Girshick, R., & Sun, J. (2017). Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks. IEEE Transactions on Pattern Analysis & Machine Intelligence, 39(06), 1137-1149.
[77] Rey, J. (2018). Faster R-CNN: Down the rabbit hole of modern object detection. Tryolabs.
[78] Schmidt, W. E., Woodward, B. T., Millikan, K. S., Guza, R. T., Raubenheimer, B., & Elgar, S. (2003). A GPS-tracked surf zone drifter. Journal of Atmospheric and Oceanic Technology, 20(7), 1069-1075.
[79] Scott, T., Masselink, G., & Russell, P. (2011). Morphodynamic characteristics and classification of beaches in England and Wales. Marine Geology, 286(1-4), 1-20.
[80] Scott, T., Austin, M., Masselink, G., & Russell, P. (2016). Dynamics of rip currents associated with groynes—field measurements, modelling and implications for beach safety. Coastal Engineering, 107, 53-69.
[81] Shepard, F. P., Emery, K. O., & La Fond, E. C. (1941). Rip currents: a process of geological importance. The Journal of Geology, 49(4), 337-369.
[82] Shepard, F. P., & Inman, D. L. (1950). Nearshore water circulation related to bottom topography and wave refraction. Eos, Transactions American Geophysical Union, 31(2), 196-212.
[83] Short, A. D. (1992). Beach systems of the central Netherlands coast: processes, morphology and structural impacts in a storm driven multi-bar system. Marine Geology, 107(1-2), 103-137.
[84] Short, A. D. (1999). Handbook of beach and shoreface morphodynamics(No. 551.468 HAN).
[85] Short, A. D., & Hogan, C. L. (1994). Rip currents and beach hazards: their impact on public safety and implications for coastal management. Journal of Coastal Research, 197-209.
[86] Shorten, C., & Khoshgoftaar, T. M. (2019). A survey on image data augmentation for deep learning. Journal of Big Data, 6(1), 1-48.
[87] Simonyan, K., & Zisserman, A. (2014). Very deep convolutional networks for large-scale image recognition. arXiv preprint arXiv:1409.1556.
[88] Smith, L. N. (2018). A disciplined approach to neural network hyper-parameters: Part 1--learning rate, batch size, momentum, and weight decay. arXiv preprint arXiv:1803.09820.
[89] Sonu, C. J. (1972). Field observation of nearshore circulation and meandering currents. Journal of Geophysical Research, 77(18), 3232-3247.
[90] Song, Z., Fu, L., Wu, J., Liu, Z., Li, R., & Cui, Y. (2019). Kiwifruit detection in field images using Faster R-CNN with VGG16. IFAC-PapersOnLine,52(30), 76-81.
[91] Stringari, C. E., Harris, D. L., & Power, H. E. (2019). A novel machine learning algorithm for tracking remotely sensed waves in the surf zone. Coastal Engineering, 147, 149-158.
[92] Sun, X., Wu, P., & Hoi, S. C. (2018). Face detection using deep learning: An improved faster RCNN approach. Neurocomputing, 299, 42-50.
[93] Takác, M., Bijral, A., Richtárik, P., & Srebro, N. (2013, May). Mini-batch primal and dual methods for SVMs. In International Conference on Machine Learning (1022-1030). PMLR.
[94] Turner, I. L., Whyte, D., Ruessink, B. G., & Ranasinghe, R. (2007). Observations of rip spacing, persistence and mobility at a long, straight coastline. Marine Geology, 236(3-4), 209-221.
[95] Viola, P., & Jones, M. J. (2004). Robust real-time face detection. International Journal of Computer Vision, 57(2), 137-154.
[96] Wilson, D. R., & Martinez, T. R. (2003). The general inefficiency of batch training for gradient descent learning. Neural Networks, 16(10), 1429-1451.
[97] Wright, L. D., & Short, A. D. (1984). Morphodynamic variability of surf zones and beaches: a synthesis. Marine Geology, 56(1-4), 93-118.
[98] Yoon, J. J. (2017). Numerical Study of Rip Current Generation at Daecheon Beach, West Coast of Korea. Journal of Coastal Research, 79 (10079), 259-263.
[99] Zhang, C., Xu, X., & Tu, D. (2018). Face detection using improved faster rcnn. arXiv preprint arXiv:1802.02142.
[100] Zhao, X., Li, W., Zhang, Y., Gulliver, T. A., Chang, S., & Feng, Z. (2016, September). A faster RCNN-based pedestrian detection system. In 2016 IEEE 84th Vehicular Technology Conference (VTC-Fall) (1-5). IEEE.