Deep visual domain adaptation and semi-supervised segmentation for understanding wave elevation using wave flume...

Area of Science:

• Oceanography and Coastal Engineering
• Computer Vision and Machine Learning

Background:

• Accurate water surface elevation estimation is critical for understanding nearshore processes.
• Traditional in-situ acoustic sensors have limitations in measuring water levels effectively.
• Ocean wave height measurement presents significant challenges in coastal environments.

Purpose of the Study:

• To develop and validate a vision-based approach for estimating water surface elevation.
• To address challenges in measuring ocean wave height directly using a novel method.
• To utilize multi-view datasets for robust water level estimation.

Main Methods:

• Proposed a visual domain adaptation method for water level estimation.
• Implemented a semi-supervised approach for extracting wave height information from long-term sequences.
• Conducted wave flume experiments using two cameras (side and top viewpoints) for data acquisition.

Main Results:

• Validated the approach by comparing estimated water elevation time series with ground-truth wave gauge data.
• Achieved high agreement with correlation coefficients of 0.98 (measurement) and 0.95 (estimation) for regular waves.
• Demonstrated strong performance for irregular waves with correlation coefficients of 0.90 (measurement) and 0.85 (estimation).

Conclusions:

• The proposed vision-based method effectively estimates water surface elevation and wave height.
• The approach overcomes limitations of traditional sensors and provides accurate measurements.
• This technique offers a promising solution for nearshore process monitoring and research.