Comparison of Assessment of Cyclone Intensity based on Deep Learning from Satellite Data

Cyclones are one of the deadliest regular catastrophes which causes colossal annihilation. Cyclones can be extremely deadly; gaining knowledge of it beforehand is helpful for organizing and making arrangements. Several approaches have been implemented in the past to track cyclones and measure their harshness after eye formation. In order to prevent the catastrophe from cyclones quickly and rapidly, tracking the path of cyclones before the eye formation is required. The foremost goal of the research is to anticipate cyclones before the eye forms. SVM algorithm, CNN model and DenseNet model have been compared based on loss and Root Mean Square Error value. The INSAT-3D dataset is trained using DenseNet169 model. The experimental outcomes of the proposed model reveal an RMSE of 2.1893 and Loss of 1.7534. The proposed methodology provides a Web-App, a User Interface that allows user to upload INSAT-3D IR Satellite Images of Cyclones to predict the intensity of the cyclone being formed and also intimate the user through a notification, which displays the details of the cyclone being formed along with its intensity.

I. Introduction

Tropical hurricanes, which instigate in the Indian Ocean basin, are usual in India every year. One of the vilest natural calamities that may cause massive chaos is a cyclone. Tropical cyclones are fierce, unexpected natural disasters that cause an excessive damage to property and human life each year. Knowledge about cyclones and other natural adversities in advance is beneficial for planning and making preparations as it is pretty risky. So, it is essential to predict such enormous shattering incidences so that the pre-emptive steps can be executed in prior that can lesser the death toll and cut expenses. Several techniques have been articulated in the earlier time to oversee cyclones and gauge the severity after eye formation. This initiative aims to prevent damage effectively and rapidly which is caused by cyclones and track its path before the eye formation, using the temporal precision and multimodality for the tropical cyclones that occurred from 2012 to 2021 provided by the INSAT-3D IR Satellite Image dataset. The collection presently comprises of 136 cyclone INFRARED photos that were taken over the Indian Ocean by the satellite “INSAT3D” with the specified intensities. In order to address the problem of cyclone assessment, several methodologies are used and compared in this study. The primary objective is to forecast cyclones beforehand the formation of the eye, which is a challenging effort that entails efficacy. Traditional approaches can be applied to spot cyclones and forestall their advancement. An imagery adopted methodology was implemented to evaluate the strength which is based on the inner construction and the analysis of typhoon, but this technique resulted in a very less accuracy. The angle difference deviation methodology was extensively utilized to estimate the intensity of the cyclone in the Pacific Ocean and in the northern part of Atlantic [14] [15]. Several other models, such as Support Vector Machine (SVM), DenseNet and Convolutional Neural Network (CNN) have been associated which normalizes the harshness of the cyclone.