The accurate prediction of tropical cyclone severity is of paramount importance in mitigating the potential damages arising from such catastrophic events. Constant monitoring and precise forecasting of tropical cyclones using remote satellite imagery from the Meteorological and Oceanographic Satellite Data Archival Centre (MOSDAC) are crucial. However, the challenge encountered with the current deep learning approach to image classification is its reliance on extensive labelled data and its limitations in few-shot learning. This study proposes a novel few-shot learning (FSL) approach for the prediction of tropical cyclone severity. In conjunction with FSL, the earth mover's distance (EMD) metric is employed to compute the distance between dense regions, thereby determining the relevance of an image. The methodology harnesses a remote satellite dataset provided by MOSDAC. The proposed approach is underpinned by the human capacity to identify novel classes from a limited number of samples, leveraging previously acquired knowledge. The FSL methodology adopts a meta-learning mechanism, enabling enhanced understanding of the data and facilitating the generalization of a new class of data. The results indicate that the FSL+EMD-based models outperform other state-of-the-art models, achieving a prediction accuracy of 85.8% in forecasting tropical cyclone severity from remote satellite imagery.