Advanced DL Based Multiscale RNN for Effectual Classification of Cyclone Prediction

Tropical cyclones are severe weather events that pose significant risks to human life and property, making accurate classification and intensity prediction essential for disaster prevention. This research introduces Deep Learning (DL) based approach for cyclone classification, integrating advanced image processing and classification techniques. A Multiscale Recurrent Neural Network (RNN) framework is proposed, the input image with Wiener filtering for noise reduction, followed by multiscale K-means clustering for precise cyclone segmentation. To extract critical features, a wavelet transform captures multi-resolution spatial and temporal patterns, which are then classified using a Multiscale RNN to categorize cyclone intensity into low, medium, and high levels. The proposed mode, developed in Python and trained on an ideal cyclone dataset, the proposed model achieves 96% classification accuracy, enhancing prediction reliability and supporting disaster preparedness.