Physics-Guided Spatiotemporal Learning for Coastal Wave Peak Period Estimation from Video

2026-06-12 · Source: cs.AI updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Coastal Oceanography · Depth: Advanced, extended

Summary

A novel Physics-Guided Deep Spatiotemporal Learning Framework is proposed for direct estimation of nearshore wave peak periods ($T_{p}$) from passive coastal video streams. This framework addresses limitations of traditional monitoring systems and existing deep learning methods by integrating automated temporal-variance-based region-of-interest (ROI) detection, multi-stage Sim-to-Real transfer learning, and physics-informed regularization. Evaluations of various spatiotemporal architectures, including transformer-based (LtViViT) and recurrent-convolutional (TinyWaveNet), revealed that LtViViT achieved the lowest instantaneous prediction RMSE of 0.7692 s. Conversely, the lightweight PtAttnCNN (TinyWaveNet) with \u03bb=5.0 demonstrated superior operational oceanographic skill (WS: 0.9811, SI: 0.0892), low parameters (0.28 M), and fast inference (0.53 ms). Physics-guided regularization significantly enhanced trend-following consistency and eliminated physically implausible predictions, while automated ROI detection improved accuracy by 27%. The framework shows promise for cost-efficient, long-term coastal wave monitoring.

Key takeaway

For ML Engineers developing coastal monitoring systems or coastal managers seeking cost-effective solutions, this framework offers a robust approach to estimate wave peak periods from video. You should consider the PtAttnCNN (TinyWaveNet) with \u03bb=5.0 for operational deployments, prioritizing its high oceanographic skill (WS: 0.9811), low parameter count (0.28 M), and sub-millisecond inference (0.53 ms). This balances physical consistency with real-time feasibility, providing a reliable alternative to traditional sensors.

Key insights

A physics-guided deep spatiotemporal learning framework accurately estimates nearshore wave peak periods from video, leveraging multi-stage transfer learning and physical constraints.

Principles

• Physics-informed regularization prevents physically implausible predictions.
• Multi-stage transfer learning mitigates labeled data scarcity.
• Automated ROI detection improves model focus and accuracy.

Method

The framework employs temporal-variance ROI detection, synthetic pre-training (Airy wave theory), noisy "silver" data pre-training, and expert "gold" data fine-tuning with a physics-guided loss function.

In practice

• Use transformer models for high instantaneous accuracy.
• Employ recurrent-convolutional models for operational stability.
• Implement physics-guided loss to ensure valid outputs.

Topics

• Coastal Wave Monitoring
• Wave Peak Period
• Physics-Guided AI
• Spatiotemporal Deep Learning
• Video Vision Transformer
• Transfer Learning

Best for: Computer Vision Engineer, AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• Physics-Guided Spatiotemporal Learning for Coastal Wave Peak Period Estimation from Video — Physics-guided deep spatiotemporal learning from video can accurately and consistently estimate coastal wave peak periods, overcoming traditional monitoring limitations.
• PeakFocus: Bridging Peak Localization and Intensity Regression via a Unified Multi-Scale Framework for Electricity Load Forecasting — PeakFocus unifies peak timing localization and intensity regression to overcome smoothing and misalignment in electricity load forecasting.
• Outperforming Self-Attention Mechanisms in Solar Irradiance Forecasting via Physics-Guided Neural Networks — Physics-guided hybrid neural networks outperform complex self-attention for GHI forecasting in high-noise environments.
• Spatio-Temporal Models in AI: Teaching AI to Understand Space and Time — Spatio-Temporal AI models learn interconnected changes across space and time, crucial for dynamic real-world predictions.
• Physics-Guided Dual Decoding and Spectral Supervision for Global 3D Hydrometeor Prediction — PredHydro-Net uses physics-guided dual decoding and spectral supervision to improve global 3D hydrometeor prediction, overcoming smoothing issues.
• Advanced Flood Prediction with Physics-Guided Deep Learning: Combining UNet, FNO, and SAR/Optical Imagery — Physics-guided deep learning improves flood prediction accuracy and consistency by integrating multi-modal remote sensing with hydrodynamic equations.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.