机械与电子2025，Vol.43Issue(10)：3-10,8.
基于深度学习的海浪参数观测方法
A Deep Learning-based Framework for Ocean Wave Parameter Estimation
摘要
Abstract
To address the limitations of traditional wave observation methods in complex backgrounds,such as poor robustness and low computational efficiency,this study proposes an improved deep learning-based wave observation algorithm.The method directly estimates disparity maps from stereo wave images,significantly enhancing processing efficiency.The proposed network architecture consists of three main modules:feature extraction,3D convolution,and disparity regression.In the feature extraction stage,an At-rous Spatial Pyramid Pooling(ASPP)module is introduced to capture multi-scale spatial features.An at-tention mechanism is further integrated to optimize feature fusion,enabling the construction of a multi-branch cost volume.A stacked encoder-decoder structure based on 3D convolution is employed to regular-ize the cost volume and extract reliable point correspondences.Experimental results on the Acqua Alta dataset demonstrate that,compared with traditional approaches,the proposed method achieves comparable reconstruction accuracy while improving computational efficiency by approximately 75％.These results highlight the method's potential for high-efficiency and high-precision wave observation,with significant theoretical and engineering implications.关键词
深度学习/视差估计/海浪观测/三维点云Key words
deep learning/disparity estimation/ocean wave observation/3D point cloud分类
计算机与自动化引用本文复制引用
张会珍,徐相龙,王立杰,高嘉伟,侯男..基于深度学习的海浪参数观测方法[J].机械与电子,2025,43(10):3-10,8.基金项目
国家自然科学基金资助项目(2021JJLH0025) （2021JJLH0025）
