| 摘要: |
| 本文基于深度学习方法,利用工业电荷耦合器件(Charge Coupled Device,CCD)摄像机拍摄的海浪图像及对应测波雷达获取的有效波高数据,开展了海浪有效波高的反演研究。为准确提取图像中的有效波高信息,对图像进行了倾斜校正,并将校正图像输入改进后的EfficientNetB7模型进行反演。反演实验结果表明,本文提出的方法较可行,对比ResNet152、InceptionV3、DenseNet264等传统卷积神经网络模型,反演精度更高。通过这一方法,本文探索并验证了深度学习技术在复杂海况下进行波高反演的潜力,为相关领域的研究提供了新的技术路径。 |
| 关键词: 有效波高反演 卷积神经网络 海浪图像 透视变换 图像回归 |
| DOI：10.11759/hykx20241211001 |
| 分类号:TP301.6 |
| 基金项目:国家基金重点项目(U2006207) |
|
| Research on sea wave height estimation methods utilizing CCD imaging |
|
DING Chen1,2, WANG Ruifu1, MENG Junmin2
|
|
1.Shandong University of Science and Technology, Qingdao 266590, China;2.First Institute of Oceanography, Ministry of Natural Resources, Qingdao 266061, China
|
| Abstract: |
| This study investigates a deep learning-based method for retrieving significant wave height (SWH) using wave images captured by an industrial charge-coupled device (CCD) camera and corresponding SWH measurements from a wave radar. To accurately extract SWH information, the images were first tilt-corrected and then input into an improved EfficientNetB7 model for inversion. Experimental results demonstrate that the proposed method is highly feasible and achieves superior inversion accuracy compared with traditional convolutional neural network models, including ResNet152, InceptionV3, and DenseNet264. This study explores and validates the potential of deep learning techniques for SWH inversion under complex sea conditions, providing a new technical pathway for related research. |
| Key words: significant wave height inversion convolutional neural networks ocean wave images perspective transformation image regression |
