A Video Dataset for Nearshore Wave Breaking Type Classification.

Area of Science:

• Coastal geomorphology
• Oceanography
• Remote sensing

Background:

• Wave breaking type is key to nearshore hydrodynamics and energy dissipation.
• Current remote sensing methods often use static images, missing dynamic breaking event evolution.
• Video analysis offers a more comprehensive approach to studying wave breaking.

Purpose of the Study:

• To introduce the first publicly available video dataset for wave breaking type classification.
• To provide a robust dataset for training and evaluating machine learning models for wave breaking analysis.
• To advance the understanding of nearshore hydrodynamic processes through dynamic video data.

Main Methods:

• Collected 9,000 labeled video clips of wave breaking from 15 cameras across diverse coastal sites.
• Implemented a rigorous data curation workflow: video segmentation, cropping, labeling, and frame extraction.
• Utilized a deep learning architecture combining Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN) for classification.

Main Results:

• The developed dataset encompasses three primary wave breaking types: Spilling, Plunging, and Surging.
• Classification experiments achieved state-of-the-art performance using the CNN-RNN deep learning model.
• The dataset's quality was ensured through a meticulous curation process.

Conclusions:

• The new video dataset significantly enhances the capability for remote sensing of wave breaking types.
• This resource facilitates more accurate modeling of wave energy dissipation and nearshore processes.
• The findings highlight the effectiveness of deep learning for analyzing dynamic coastal phenomena.