Phy-CoCo

The code of "Phy-CoCo: Physical Constraint-based Correlation Learning for Tropical Cyclone Intensity and Size Estimation" accepted by ECAI2024.

Introduction

Contribution:

We proposed CoM based on Centrally Expanded Pooling (CEP) to model the correlation between the extracted features and the estimated attributes, fully exploring task-specific features.
To facilitate cross-task interaction, we designed bidirectional physical constraints applied to the transformation of features of interrelated tasks using Multi-Domain Recurrent Convolutions (MDRC).
Extensive experiments are conducted on multi-modal TC datasets to demonstrate the superiority of Phy-CoCo over the state-of-the-art TC estimation methods. The results highlight that Phy-CoCo is effective for both TC MSW and RMW estimation.