Cyclone Intensity Detection and Classification Using a Attention-Based 3D Deep Learning Model

Abstract

To investigate the feasibility of determining tropical cyclone (TC) intensity using satellite photos, a deep learning convolutional neural network system is employed. This paper also demonstrates how to determine a cyclone’s strength using an image processing method. To identify cyclone components, a convolutional neural network-based method is employed. Predicting and determining the intensity of flash floods are essential to lowering the number of fatalities and property damage caused by tropical cyclones (TC). This study describes a method for evaluating TC strength from satellite images. The convolutional neural network model is utilised for cyclone detection and characterisation, whilst AlexNet is used to extract features, build models, and forecast cyclone conditions. The findings imply that our model qualitatively developed a perspective resembling that of subject-matter specialists. Additionally, we use the convolutional block attention module (CBAM), which is based on 3D, to simulate visual attention in order to improve the model’s focus on the crucial channels and primary cloud structure. According to the analysis from the study, the suggested model’s root mean square error (RMSE) is 9.5 kts, which is lESS than both the classic deep learning (DL) method of intensity estimate and the advanced Dvorak technique (ADT) by 9.3% and 27%, respectively.