Abstract
Classification of weather images is very significant in climate monitoring, disaster management, and weather forecasting. This work is an innovation on the architecture that combines U-Net with a BiLSTM network for classification accuracy improvement. U-Net is one of the most advanced convolutional neural networks, and it is characterized by fine-grained feature extraction and spatial information preservation in images. The model combines it with BiLSTM, which focuses on detecting temporal and contextual patterns, thereby enabling a combined understanding of the spatial and temporal dimensions of weather phenomena. It strengthens all aspects: in addition to how the spatial U-Net's capabilities build strong arguments and why the recognition of correlations with a model such as BiLSTM significantly improves this method against more common, traditionally used, models of the convolutional neural networks family. Tests, also carried out, showed excellent and improved parameters during evaluation like precision, recall, and balance measures, when evaluating results of similar common datasets commonly used within weather forecast systems analysis. It provides a strong structure for enhancing the classification of weather images and supporting the study of meteorology by bringing in the strengths of convolutional networks and recurrent networks.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call