Measuring the rogue wave pattern triggered from Gaussian perturbations by deep learning.

Area of Science:

• Nonlinear physics
• Fluid dynamics
• Computer vision

Background:

• Modulational instability can generate rogue waves (RWs) from weak Gaussian perturbations on plane wave backgrounds.
• Numerical simulations suggest these RWs exhibit similar structural patterns.
• Automatic measurement of RW patterns has been a significant challenge, hindering empirical validation.

Purpose of the Study:

• To automatically detect and analyze rogue wave patterns using computer vision techniques.
• To validate the hypothesis that rogue waves generated from different perturbations share similar visual characteristics.
• To introduce a novel deep learning model and dataset for rogue wave analysis.

Main Methods:

• Development of a deep neural network model named rogue wave detection network (RWD-Net) for automated RW detection.
• Creation and release of the rogue wave dataset-10K (RWD-10K), comprising 10,191 annotated RW images.
• Application of RWD-Net to the RWD-10K dataset for performance evaluation.

Main Results:

• The RWD-Net achieved an average precision of 99.29% on the test set of the RWD-10K dataset.
• The model's high accuracy provides strong evidence for the similar computer vision patterns of rogue waves.
• A new metric, the density of RW units, was derived to characterize perturbation evolution.

Conclusions:

• Deep learning models, like RWD-Net, can effectively automate the detection of rogue waves.
• The findings support the existence of consistent visual patterns across different rogue wave instances.
• The RWD-10K dataset and RWD-Net offer valuable tools for future research in nonlinear physics and fluid dynamics.