Deep Learning Based Cyclone Intensity Estimation

Abstract

Background/Objectives: Cyclones are perilous, and disaster management depends heavily on accurate intensity estimation. The primary objective of this research is to develop a deep learning-based technique that uses satellite imagery to assess cyclone intensity. The research involves use of Capsule Convolutional Neural Networks (Caps Nets), which are intended to handle complex patterns and preserve spatial hierarchies. With the ability to preserve positional information and enhance the perception of spatial relationships in satellite imagery, Caps Nets overcome certain drawbacks of conventional Convolutional Neural Networks. Regression modeling was used in this study to forecast cyclone strength based on INSAT 3D satellite imagery, offering a structured technique for evaluating meteorological data and producing precise forecast estimations. The results indicate that the Caps Nets-based approach yields more accurate cyclone intensity estimates compared to traditional Convolutional Neural Networks with a MSE of 1.51 and RMSE of 2.07. The improved pattern identification capabilities of this method contributes to improved estimates of wind speed and evaluations of cyclone intensity. The findings indicate that this technique has the potential to considerably improve disaster response and preparedness by providing more precise and accurate data for evacuation planning and early warning systems. By retaining spatial hierarchies and comprehending complex patterns in satellite imagery, Capsule Convolutional Neural Network offers a novel approach to cyclone intensity estimation, providing improved accuracy and robustness compared to conventional methods.