Objective wave breaker type classification from flume video using 3D CNNs across multiple plane slopes

Wave breaking is a complex and energy-intensive process in coastal hydrodynamics, yet breaker type classification remains largely subjective and discontinuous. This study enables more objective and consistent classification of breaker types through the Wave Breaker Video Classifier (WaBViC), a machine-learning framework. WaBViC integrates spatiotemporal visual features with experimentally controlled parameters, enabling data-driven predictions within the experimental range covered by the dataset. A balanced dataset of breaker events obtained from controlled flume experiments with three impermeable slope configurations was used for training and validation. Results show that incorporating experimentally controlled parameters improves classification accuracy and confines errors to neighboring types, while the probabilistic output preserves the continuity of breaking transitions. Additional analyses indicated that classification reliability is affected by image quality and viewing geometry, underscoring the need for refined image rectification and parallax-robust feature extraction. Comparison with the χ- and ξ 0 -based surf-similarity frameworks demonstrated that the data-driven classifications reproduce established steepness–depth and steepness–slope relationships while suggesting that their relative weighting may require re-evaluation. These findings show that breaker type classification can be achieved through a reproducible machine-learning-assisted procedure. This study thus provides a data-driven basis for objective wave breaking analysis and a reference for future work on coastal hydrodynamics and energy-dissipation modeling.