Classification of tropical cyclones observed by satellites in the Southwest Indian Ocean basin using the 2D-CNN model

The aim of this work was to model the images of tropical cyclones seen from the meteorological satellites using the 2D-CNN model. The work remained within the framework of the classification problem. More specifically, the experiment consisted of training, then testing and validating a 2D-CNN model that predicts the class of the cyclone image using a set of images of tropical cyclones obtained by satellites. The model only took into account 2 classes. The first (respectively the second) class contains images of cyclones with a wind force less than 34Kt (respectively greater than or equal to 34Kt). In fact, the cyclone begins to be named as soon as the force of the wind that accompanies it is greater than or equal to 34Kt. The simulations were done for each value of the batch size corresponding to the power of 2, namely 2, 4, 8, 16, 32, 64, 128 and 254. For each of these batch size values, the simulations were evaluated using the evolution of the mean curve between training and test data as a function of the epoch. The best classification model was obtained by running the simulation with a batch size of 64 at the 25th epoch with a model reliability rate of 97%.