海洋波浪監測技術用於測量與記錄海洋表面波動的波高、週期及波向等數值。隨著人工智慧(AI)領域的發展，人工智慧模型開始具備處理非線性迴歸(Regression)問題的能力，將其應用於波浪監測系統也成為可能。此外，光學(Optic)攝影站具有成本低、建置快速等優點，因此本研究以深度學習技術建置以海面光學動態影像即時辨識波高之模型，並探討海面影像之時間相依性與特徵尺度對波高監測之重要性，以此實現示性波高的即時估算(Real-time Estimation)。
長期遞迴卷積網路(Long-term Recurrent Convolutional Network, LRCN)融合卷積神經網路(Convolutional Neural Network, CNN)與長短期記憶網路(Long Short Term Memory Network, LSTM)，能處理隨空間、時間變動的數據。本研究以長期遞迴卷積網路建置波高辨識模型，同時以海岸光學攝影機及海氣象資料浮標做為光學動態影像及波高數據來源。比較使用五種卷積神經網路架構及多尺度特徵融合技術(Multi-Scale Feature Fusion)之模型效能，研究結果表明，不使用多尺度特徵融合技術之VGG19卷積神經網路架構表現最佳，過深的卷積神經網路容易導致模型退化與過度擬合的問題，而引入影像之多尺度特徵將增加學習數據中的雜訊，使模型對示性波高的判斷有較大的誤差。本研究也針對模型中長短期記憶網路之架構進行調整，研究結果顯示以使用Stacked LSTM之模型驗證誤差最小。此外，以不同時長之動態影像訓練模型，誤差將隨時長增長而減少，時長超過八秒誤差變化趨緩。此結果表明波浪影像序列具有時間相依性，透過觀察其在時間上的變化可以對示性波高進行更準確的估計。
經過測試，本研究建置之即時波高辨識模型，模型計算之波高數值與實測波高數值之相關係數可達0.88，表示模型具有優秀的波高辨識能力，模型辨識速度平均每筆波高數據耗時1.5秒，具有即時運算的能力。並且與使用單張影像之模型進行比較，發現使用動態影像之模型具有較準確的辨識結果，顯示考慮海面在時間上的變化對於波高辨識的重要性。再與前人研究結果比較，發現本研究使用之模型透過觀察時長較短的動態影像即可達到與前人相近的準確率。以上結果均顯示此類技術應用在光學影像辨識波高可提供快速、準確的觀測結果。最後為評估模型的泛化能力，將模型應用在三台不同拍攝地點的海岸攝影機影像，辨識結果之無因次化均方根誤差均在三成以內，顯示將本研究之模型應用在其他場域也具有辨識當地浪況的能力。

The development of AI and optical technology have enhanced the possibility of integrating artificial intelligence with wave observation systems. This study employs deep learning to develop a real-time wave height recognition model based on sea surface video.
In this study, an LRCN-based wave height recognition model was developed, utilizing coastal optical camera and data buoy as data sources. The Long-term Recurrent Convolutional Network (LRCN), which combines CNN and LSTM, can process 3-dimensional data. The performance of models considering five CNN architectures and Multi-Scale Feature Fusion techniques were compared. The results indicated that the VGG19 architecture without Multi-Scale Feature Fusion performed best. Deeper architecture tends to cause model degradation and overfitting, and introducing multi-scale features in the images increases noise in the training data. The LSTM architecture adjustment was also considered, revealing that the model using Stacked LSTM had the lowest error. Additionally, the model validation error decreased with longer video durations, indicates that wave image sequences have temporal dependencies. The model was applied to images from three different coastal cameras to evaluate the generalization ability. The normalized RMSE was within 30%, indicating that the model can also recognize local wave conditions when applied to other fields.
In test stage, the wave height recognition model achieved a correlation coefficient of 0.88. Compared with models using images, the model using image sequence provided more accurate recognition results.

[1] 楊大成、范揚洺、李汴軍、高家俊、莊士賢、 滕春慈(2010)，氣候變遷對臺灣東北角海域波候變化之初步研究，第32屆海洋工程研討會論文集，第155-160頁。
[2] 交通部中央氣象署(2019)，異常海象機率預警研究與作業試用，研究計畫報告。
[3] 葉浩君(2020)，臺灣鄰近海域湧浪研究。國立成功大學碩士論文。
[4] 吳立中、董東璟、滕春慈、吳益裕(2021)，臺灣海域作業化海氣象資料浮標監測網 ，海洋及水下科技季刊，第三十一卷，第三期，第9-14頁。
[5] 王敘民、董東璟、蔡政翰、林芳如(2022)，海岸裂流觀測之研究，中國土木水利工程學刊，第四十九卷，第六期，第24-30頁。
[6] 交通部運輸研究所(2023)，應用影像智慧化技術判釋海岸公路及防波堤越波研究(1/4)-日間越波影像判釋，研究計畫報告。
[7] Buscombe, D., Carini, R. J., Harrison, S. R., Chickadel, C. C., and Warrick, J. A. (2020). Optical wave gauging using deep neural networks. Coastal Engineering, 155, 103593.
[8] Chellapilla, K., Puri, S., and Simard P. (2006). High Performance Convolu-tional Neural Networks for Document Processing, in 10th Interna-tional Workshop on Frontiers in Handwriting Recognition, (La Baule(France)), Universit ́e de Rennes 1, Suvisoft.
[9] Donahue, J., Hendricks, L. A., Rohrbach, M., Venugopalan, S., Guadarrama, S., Saenko, K., & Darrell, T. (2017). Long-Term Recurrent Convolutional Networks for Visual Recognition and Description. IEEE Transactions on Pattern Analysis and Machine Intelligence, 39(4), 677–691.
[10] Ellenson, A. N., Simmons, J. A., Wilson, G. W., Hesser, T. J., and Splinter, K. D. (2020). Beach State Recognition Using Argus Imagery and Convolutional Neural Networks. Remote Sensing, 12(23), 3953.
[11] Graves, A. and Schmidhuber, J. (2005). Framewise phoneme classification with bidirectional LSTM and other neural network architectures. Neural networks, 18(5-6), 602-610.
[12] Graves, A., Mohamed, A. R., & Hinton, G. (2013). Speech recognition with deep recurrent neural networks. In 2013 IEEE international conference on acoustics, speech and signal processing, 6645-6649.
[13] Haghshenas, A., Hasan, A., Osen, O., and Mikalsen, E. T. (2023). Predictive digital twin for offshore wind farms. Energy Informatics, 6(1).
[14] He, K., Zhang, X., Ren, S., & Sun, J. (2016). Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, 770-778.
[15] Hochreiter, S., and Schmidhuber, J. (1997). Long short-term memory. Neural computation, 9(8), 1735-1780.
[16] Horn, B. K., and Schunck, B. G. (1981). Determining optical flow. Artificial Intelligence, 17(1–3), 185–203.
[17] Huang, W., Liu, X., and Gill, E. (2017). Ocean Wind and Wave Measurements Using X-Band Marine Radar: A Comprehensive Review. Remote Sensing, 9(12), 1261.
[18] Hwang, I. K., Lee, M., Han, J., and Choi, J. (2023). Wave height measurement scheme using wave detector based on convolutional neural network and PPM calculator with ocean wave images. International Journal of Naval Architecture and Ocean Engineering, 15, 100542.
[19] Kang, B., and Duran Vinent, O. (2023). The Application of CNN-Based Image Segmentation for Tracking Coastal Erosion and Post-Storm Recovery. Remote Sensing, 15(14), 3485.
[20] Kim, Y. H., Cho, S., and Lee, P. S. (2023). Wave height classification via deep learning using monoscopic ocean videos. Ocean Engineering, 288, 116002.
[21] Lecun, Y., Bottou, L., Bengio, Y. and Haffner, P. (1998). Gradient-based learning applied to document recognition. Proceedings of the IEEE, 86(11), 2278–2324.
[22] Li, T., Hua, M. and Wu, X. (2020). A Hybrid CNN-LSTM Model for Forecasting Particulate Matter (PM2.5), in IEEE Access, vol. 8, 26933-26940
[23] Li, J., Kong, X., Yang, Y., Yang, Z. and Hu, J. (2022). Optical flow based measurement of flow field in wave-structure interaction. Ocean Engineering, 263, 112336.
[24] 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, 2117-2125.
[25] Liu, W., Anguelov, D., Erhan, D., Szegedy, C., Reed, S., Fu C. and Berg, A.C. (2016) SSD: Single Shot MultiBox Detector. EuropeanConference on Computer Vision, 2016, 21-37.
[26] Mahjoobi, J., and Adeli Mosabbeb, E. (2009). Prediction of significant wave height using regressive support vector machines. Ocean Engineering, 36(5), 339–347.
[27] Mutegeki, R. and Han, D. S. (2020). A CNN-LSTM Approach to Human Activity Recognition, 2020 International Conference on Artificial Intelligence in Information and Communication (ICAIIC), Fukuoka, Japan, 2020, 362-366.
[28] Scardino, G., Scicchitano, G., Chirivì, M., Costa, P. J. M., Luparelli, A., and Mastronuzzi, G. (2022). Convolutional Neural Network and Optical Flow for the Assessment of Wave and Tide Parameters from Video Analysis (LEUCOTEA): An Innovative Tool for Coastal Monitoring. Remote Sensing, 14(13), 2994.
[29] Seemann, J., Senet, C. M., Dankert, H., Hatten, H., & Ziemer, F. (1999). Radar image sequence analysis of inhomogeneous water surfaces. In Applications of Digital Image Processing XXII, 3808, 536-546, SPIE.
[30] Simonyan, K., & Zisserman, A. (2014). Very deep convolutional networks for large-scale image recognition. arXiv preprint arXiv:1409.1556.
[31] Tibshirani, R. (1996). Regression shrinkage and selection via the lasso. Journal of the Royal Statistical Society Series B: Statistical Methodology, 58(1), 267-288.
[32] Vousdoukas, M. I., Ferreira, P. M., Almeida, L. P., Dodet, G., Psaros, F., Andriolo, U., Taborda, R., Silva, A. N., Ruano, A. and Ferreira, S. M. (2011). Performance of intertidal topography video monitoring of a meso-tidal reflective beach in South Portugal. Ocean Dynamics, 61(10), 1521–1540.
[33] Wang, L., Deng, X., Ge, P., Dong, C., J. Bethel, B., Yang, L. and Xia, J. (2022). CNN-BiLSTM-Attention Model in Forecasting Wave Height over South-East China Seas. Computers, Materials and Continua, 73(1), 2151–2168.
[34] Young, I. R., Rosenthal, W., Ziemer, F. (1985). A three-dimensional analysis of marine radar images for the determination of ocean wave directionality and surface currents. J. Geophys. Res. 90, 1049–1059.
[35] Zhang, X., Li, Y., Gao, S., and Ren, P. (2021). Ocean Wave Height Series Prediction with Numerical Long Short-Term Memory. Journal of Marine Science and Engineering, 9(5), 514.