Session: 15-09-01 Marine Environment and Offshore Structure Design under Climate Changes

Submission Number: 181296

Orbit: Semi-Supervised Virtual–Real Bridging for Robust Ocean Wave Direction Estimation from Pose-Aware Temporal Imagery 

Accurate estimation of ocean wave direction is essential for ship motion prediction, offshore structure operation, and ensuring maritime safety. However, developing robust estimation models remains challenging due to the scarcity of labeled real-sea data and the significant domain gap between simulated and real-world ocean environments. To address these limitations, this study introduces Oceans Real–sim Bridging for dIrecTion (ORBIT), a novel semi-supervised learning framework that effectively unifies simulated and real observational data for robust and scalable wave direction estimation. ORBIT employs a consistency-based training paradigm, where high-confidence pseudo-labels derived from weakly augmented samples are used as supervisory signals under stronger augmentations, allowing the model to leverage large volumes of unlabeled real-sea imagery. The network is designed to output unit-vector representations of wave directions, preserving angular continuity and reducing ambiguity in circular regression. Furthermore, temporal fusion across consecutive frames is incorporated to enhance prediction stability and robustness under dynamically changing sea states. Extensive experiments conducted on both simulated datasets and real-sea video imagery demonstrate that ORBIT substantially outperforms conventional fully supervised models trained with limited labeled data, achieving higher directional accuracy, improved temporal consistency, and better generalization to unseen environments. These results highlight the potential of virtual–real domain bridging and semi-supervised learning to enable practical, vision-based ocean wave sensing systems for digital twin modeling, autonomous maritime operations, and intelligent offshore monitoring in future ocean engineering applications.

Presenting Author: Runtao Zhang Shanghai Jiao Tong University

Presenting Author Biography: He is a Phd student at Shanghai Jiao Tong University.

Authors:

Runtao Zhang Shanghai Jiao Tong University
Xinde Yang Shanghai Jiao Tong University
Hongdong Wang Shanghai Jiao Tong University
Mingyang Zhang Shanghai Jiao Tong University