裂流是一種好發於近岸的強勁離岸流，其流速可達每秒 2 公尺，對於從 事海域遊憩活動的民眾有相當大的威脅。台灣四面環海，根據統計全台灣約 有 43 處海灘為裂流潛勢區域。為了提升海域遊憩活動的安全性，建立裂流警 報機制十分重要。由於以肉眼辨識裂流有一定的難度，因此，本研究利用 YOLOv7 物件偵測深度學習演算法建立裂流辨識模型，並且探討各種最佳化 方法對模型效能的影響。
本研究使用設置於外澳飛行傘基地的海岸監測攝影機拍攝之裂流案例製 作資料集，並以遷移式學習(Transfer Learning)的方式建立模型。為了強化影 像中的裂流特徵，本研究使用時間平均影像搭配一系列的影像擴增(Data Augmentation)最佳化模型，模型的精確率可達到 97.0%;召回率可達到 95.9%。 結果顯示使用時間平均影像可建置效能卓越的裂流辨識模型。
為了實現即時的裂流警報機制，本研究擬將模型佈署至 Nvidia Jetson Orin Nano 邊緣運算裝置。邊緣運算裝置為一種成本較低、體積小、低耗能的硬體， 然而其運算效能及儲存空間不如一般商業規格電腦強大。為了克服硬體效能 受限的問題，本研究使用模型壓縮(Model Compression)方法，對裂流模型進行 結構化模型剪枝(Structured Model Pruning)以降低模型的運算量。結果顯示相 較於沒有剪枝過的模型，剪枝率 90%的模型在電腦上運行時辨識速率提升約 19%，且其辨識能力維持一定水準，精確度仍可達近九成，召回率仍維持在 九成五以上。在剪枝率 95%時，模型始可在邊緣運算裝置上執行，辨識速率 可達 39 FPS。結果顯示模型經過壓縮後仍可維持良好的辨識能力，有助於裂 流警報機制的建立。

Rip currents are a type of strong offshore flow commonly occurring near the shore, with speeds reaching up to 2 meters per second, posing a significant threat to tourists engaging in marine recreational activities. Taiwan, being surrounded by the sea, has approximately 43 beaches identified as potential rip current zones according to statistics. To enhance the safety of marine recreational activities, establishing a rip current warning mechanism is crucial. Since identifying rip currents with the human eye is challenging, this study utilizes the YOLOv7 object detection deep learning algorithm to develop a rip current recognition model and explores the impact of various optimization methods on the model’s performance.
This study employs a dataset created from rip current cases captured by coastal monitoring cameras installed at the Waiou Paragliding Base, using transfer learning to develop the model. To enhance the features of rip currents in the images, this study uses time averaged images combined with a series of data augmentation techniques to optimize the model. The model achieves an accuracy of 97.0% and a recall rate of 95.9%.
To implement a real-time rip current warning mechanism, this study plans to deploy the model on the Nvidia Jetson Orin Nano edge device. Edge devices are cost-effective, compact, and low-power hardware, but their computational performance and storage capacity are not as robust as normal commercial computers. To overcome the limitations of the hardware, this study employs model compression methods, specifically structured model pruning, to reduce the computational load of the rip current model. The results show that, compared to the unpruned model, the model with a 90% pruning rate achieved approximately a 19% increase in recognition speed when run on a computer, while maintaining a high level of recognition ability, with an accuracy close to 90% and a recall rate above 95%. At a pruning rate of 95%, the model could run on the edge computing device, achieving a recognition speed of 39 FPS. The results demonstrate that the compressed model can still maintain good recognition capabilities, contributing to the establishment of a rip current warning mechanism.

[1] 王敘民，邱啓敏，梁立翰，董東璟(2020)，應用 SCHISM 模式模擬裂流時空 分布，第 42 屆海洋工程研討會論文集 350-355。
[2] 王敘民，邱啟敏，董東璟，蔡政翰(2021)，裂流模擬與影響因子之評析，第 43 屆海洋工程研討會論文集 349-354。
[3] 王敘民，董東璟，蔡政翰，林芳如(2022)，海岸裂流觀測之研究，《土木水利》 49 卷 6 期 Pp. 24-30
[4] 王敘民，董東璟，蔡政翰，滕春慈(2022)，以數值模式探討裂流生成之海 氣象條件以台南漁光島海灘為例，第 44 屆海洋工程研討會論文集。
[5] 王敘民，林芳如，滕春慈，董東璟，蔡政翰(2023)，應用 SCHISM 模式模 擬不同地形對裂流之影響，第 45 屆海洋工程研討會摘要集。
[6] 林雪美(2007)，臺灣北海岸海灘安全性之地形動力學研究，2007 年地理學者 學術研討會。
[7] 林雪美(2009)，宜蘭外澳海灘裂流判釋與遊憩安全告示牌設計，交通部觀光 局計畫。
[8] 林雪美(2009) 台灣東北部福隆海灘類型和裂流分布之長期變動研究，地理 研究(50):47-65。
[9] 郭平巧，許弘莒，張裕弦，劉景毅，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 453-457。
[10] 傅仲偉(2022)，利用深度學習之更快速區域卷積神經網路分析裂流影像。
[11] 蔡政翰，董東璟(2022)，海岸裂流監測與預警技術研究，交通部中央氣象局委託研究計畫(期末)成果報告。
[12] 蕭仁豪，溫志中，李宗霖，蕭坤欣(2014)，海水域場海灘安全性評估研究，第 36 屆 海洋工程研討會論文集 141-146。
[13] Benedet, L., Finkl, C. W., and Klein, A.H.F. 2004. Morphodynamic classification of beaches on the Atlantic coast of Florida: Geographical variability of beach types, beach safety and coastal hazards. Journal of Coastal Research, SI(39).
[14] Bochkovskiy, A., Wang, C. Y., & Liao, H. Y. M. (2020). Yolov4: Optimal speed and accuracy of object detection. arXiv preprint arXiv:2004.10934.
[15] Brander, R. W., Drozdzewski, D., & Dominey-Howes, D. (2014). “Dye in the Water” A Visual Approach to Communicating the Rip Current Hazard. Science Communication, 36(6), 802-810.
[16] Cao, K., Liu, Y., Meng, G., & Sun, Q. (2020). An overview on edge computing research. IEEE access, 8, 85714-85728.
[17] Chen, S. T., & Wang, Y. W. (2020). Improving coastal ocean wave height forecasting during typhoons by using local meteorological and neighboring wave data in support vector regression models.Journal of Marine Science and Engineering, 8(3), 149.
[18] 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).
[19] 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.
[20] 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.
[21] 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).
[22] Elgar, S., Raubenheimer, B., & Guza, R. T. (2001). Current meter performance in the surf zone. Journal of Atmospheric and Oceanic Technology, 18(10), 1735-1746.
[23] Felzenszwalb, P., McAllester, D., & Ramanan, D. (2008, June). A discriminatively trained, multiscale, deformable part model. In 2008 IEEE conference on computer vision and pattern recognition (pp. 1-8). Ieee.
[24] Frankle, J., & Carbin, M. (2018). The lottery ticket hypothesis: Finding sparse, trainable neural networks. arXiv preprint arXiv:1803.03635.
[25] Girshick, R. (2015). Fast r-cnn. In Proceedings of the IEEE international conference on computer vision (pp. 1440-1448).
[26] 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 (pp. 580-587).
[27] 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.
[28] Hass, F. S., & Jokar Arsanjani, J. (2020). Deep learning for detecting and classifying ocean objects: application of YoloV3 for iceberg–ship discrimination. ISPRS International Journal of Geo-Information, 9(12), 758.
[29] He, K., Zhang, X., Ren, S., & Sun, J. (2015). Spatial pyramid pooling in deep convolutional networks for visual recognition. IEEE transactions on pattern analysis and machine intelligence, 37(9), 1904-1916.
[30] Hochreiter, S., & Schmidhuber, J. (1997). Long short-term memory. Neural computation, 9(8), 1735-1780.
[31] Holland, K. T., Holman, R. A., Lippmann, T. C., Stanley, J., & Plant, N. (1997). Practical use of video imagery in nearshore oceanographic field studies. IEEE Journal of oceanic engineering, 22(1), 81-92. 71
[32] Holman, R. A., & Stanley, J. (2007). The history and technical capabilities of Argus. Coastal engineering, 54(6-7), 477-491.
[33] Hsu-Min Wang., Dong-Jiing Doong., Cheng-Han Tsai., Fang-Ru Lin. (2024). Rip current simulation by various meteo-ocean and terrain conditions using SCHISM model. The 11th East Asian Workshop for Marine Environment and Energy (EAWOMEN2 2023), 67-68.
[34] Huynh, L. N., Lee, Y., & Balan, R. K. (2017, June). Deepmon: Mobile gpu-based deep learning framework for continuous vision applications. In Proceedings of the 15th Annual International Conference on Mobile Systems, Applications, and Services (pp. 82-95).
[35]Inch, K. (2014). Surf zone hydrodynamics: Measuring waves and currents. Geomorphological techniques, 3, 1-13.
[36] Johnson, D., & Pattiaratchi, C. (2004). Transient rip currents and nearshore circulation on a swell‐dominated beach.Journal of Geophysical Research: Oceans, 109(C2).
[37] Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2017). ImageNet classification with deep convolutional neural networks. Communications of the ACM, 60(6), 84-90.
[38] LeCun, Y., Bottou, L., Bengio, Y., & Haffner, P. (1998). Gradient-based learning applied to document recognition. Proceedings of the IEEE, 86(11), 2278-2324.
[39] 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).
[40] Lin, T. Y., Maire, M., Belongie, S., Hays, J., Perona, P., Ramanan, D., ... & Zitnick, C. L. (2014). Microsoft coco: Common objects in context. In Computer Vision– ECCV 2014: 13th European Conference, Zurich, Switzerland, September 6-12, 2014, Proceedings, Part V 13 (pp. 740-755). Springer International Publishing. 72
[41] Liu, J., Yu, C., Hu, Z., Zhao, Y., Bai, Y., Xie, M., & Luo, J. (2020). Accurate prediction scheme of water quality in smart mariculture with deep Bi-S-SRU learning network. IEEE Access, 8, 24784-24798.
[42] Liu, Y., & Wu, C. H. (2019). Lifeguarding Operational Camera Kiosk System (LOCKS) for flash rip warning: Development and application.Coastal engineering, 152, 103537.
[43] Longuet-Higgins, M.S., and Stewart, R.W. 1964. Radiation Stresses in Water Waves; A Physical Discussion, with Applications[J]. Deep Sea Research and Oceanographic Abstracts, 11(4): 529-562.
[44] Lushine, B.J. 1991. A study of rip current drownings and related weather factors. Nation Weather Service Forecast Office, 13-19.
[45] 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.
[46] McCulloch, W. S., & Pitts, W. (1943). A logical calculus of the ideas immanent in nervous activity. The bulletin of mathematical biophysics, 5, 115-133.
[47] Meadows, G. A., Grimm, A., Brooks, C. N., & Shuchman, R. A. (2015). Remote sensing-based detection and monitoring of dangerous nearshore currents.
[48] Meng, Z. F., Chen, Z., Khoo, B. C., & Zhang, A. M. (2022). Long-time prediction of sea wave trains by LSTM machine learning method. Ocean Engineering, 262, 112213.
[49] Pirhooshyaran, M., & Snyder, L. V. (2020). Forecasting, hindcasting and feature selection of ocean waves via recurrent and sequence-to-sequence networks. Ocean Engineering, 207, 107424.
[50] 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), 912-916.
[51] Pritchard, D. W., & Carpenter, J. H. (1960). Measurements of turbulent diffusion in estuarine and inshore waters. Hydrological Sciences Journal, 5(4), 37-50.
[52] 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 (pp. 779-788).
[53] Redmon, J., & Farhadi, A. (2017). YOLO9000: better, faster, stronger. InProceedings of the IEEE conference on computer vision and pattern recognition (pp. 7263-7271).
[54] Redmon, J., & Farhadi, A. (2018). Yolov3: An incremental improvement. arXiv preprint arXiv:1804.02767.
[55] Ren, S., He, K., Girshick, R., & Sun, J. (2015). Faster r-cnn: Towards real-time object detection with region proposal networks. Advances in neural information processing systems, 28.
[56] Satyanarayanan, M. (2017). The emergence of edge computing. Computer, 50(1), 30-39.
[57] Short, A.D., and Hogan, C.L. 1994. Rip currents and Beach Hazards: Their Impact on Public Safety and Implications for Coastal Management. Coastal Res, 12: 197- 209.
[58] Shu-Ming Wang. (2021). Preliminary Study on Rip Current Observation and Simulation in Taiwan, The 10th East Asian Workshop for Marine Environment and Energy.
[59] Voulodimos, A., Doulamis, N., Doulamis, A., & Protopapadakis, E. (2018). Deep learning for computer vision: A brief review. Computational intelligence and neuroscience, 2018.
[60] Wang, C. Y., Bochkovskiy, A., & Liao, H. Y. M. (2023). YOLOv7: Trainable bag- of-freebies sets new state-of-the-art for real-time object detectors. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition (pp. 7464-7475).
[61] Yao, S., Zhao, Y., Zhang, A., Su, L., & Abdelzaher, T. (2017, November). Deepiot: Compressing deep neural network structures for sensing systems with a compressor-critic framework. In Proceedings of the 15th ACM conference on embedded network sensor systems (pp. 1-14).
[62] Yi, W., & Wang, B. (2023). Research on Underwater small target Detection Algorithm based on improved YOLOv7. IEEE Access.
[63] Zhang, K.; Zhang, M.C.; Leatherman, S.B., and Leatherman, S.P., 2020. Rip current hazard analysis. In: Malvárez, G. and Navas, F. (eds.), Global Coastal Issues of 2020. Journal of Coastal Research, Special Issue No. 95, pp. 969–973. Coconut Creek (Florida), ISSN 0749-0208.
[64] Zhao, Z., Barijough, K. M., & Gerstlauer, A. (2018). Deepthings: Distributed adaptive deep learning inference on resource-constrained iot edge clusters. IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, 37(11), 2348-2359.
[65] Zhu, D., Qi, R., Hu, P., Su, Q., Qin, X., & Li, Z. (2022). YOLO-Rip: A modified lightweight network for Rip currents detection. Frontiers in Marine Science, 9, 930478.