An Ensemble Framework for Cyclone Detection Using Satellite and Interpolated Images

Tropical cyclones, also known as hurricanes and typhoons, are considered powerful atmospheric vortices that originate over warm tropical oceans, causing severe destruction to life and property. Therefore, the early and precise detection of cyclones is crucial for effective disaster management. This study introduces a convolutional neural network-based framework for cyclone detection. The cyclone detection is performed by utilizing an interpolation technique that employs a weighted average of optical flow components. The motivation of this work is to enhance temporal resolution with the help of interpolated images derived from fractional-order optical flow. The optical flow is estimated through a fractional-order variational model designed to preserve dynamic discontinuities in cyclone motion. Optical flow helps to locate image intensity values at intermediate time steps. Furthermore, a novel convolutional neural network architecture featuring batch normalization and dropout layers is used for cyclone detection. Various accuracy metrics are employed to assess performance and compare it with existing methods.