Comparison of Hybrid Deep Learning Model for Cyclone Prediction from Raw Satellite Images

Background: Satellite image classification plays a crucial role in weather forecasting and climate analysis. It has been a focal point of attention for remote sensing communities for decades, particularly in cyclone classification. Deep Learning (DL) techniques have shown promising results in image processing. Proper cyclone identification helps to avoid damage to society and humans.Methods: In this study, we propose a method to classify cyclone types based on raw satellite images. To achieve reliable results, we propose hybrid DL models, including Convolutional Neural Network - Long Short-Term Memory (CNN-LSTM), Convolutional Neural Network - Bidirectional Long Short-Term Memory (CNN-BiLSTM), and Convolutional Neural Network - Gated Recurrent Unit (CNN-GRU). We utilized publicly available satellite images from Kaggle for model training and validation. Raw images underwent various processing techniques to boost the performance of the hybrid DL models.Results: Three hybrid DL models were tested using 50 satellite images, and performance metrics were utilized for evaluation. Experimental results show that the proposed CNN-BiLSTM model outperforms CNN-LSTM and CNN-GRU, achieving an accuracy, precision, recall, and F1 score of 98%, 99%, 98%, and 99%, respectively.Conclusion: The proposed CNN-BiLSTM model shows superior performance in cyclone classification. The outcome of the proposed model proves that it will be very useful for effective weather forecasting and climate analysis applications.