Cyclones are major weather events that often result in widespread destruction and significant loss of life. Being able to evaluate their intensity at the right time is essential for reducing the severity of their impacts. Over the past years, deep learning techniques have been progressively incorporated for analysing large collections of satellite images. Proposed methodology, present a deep learning techniques for determining cyclone intensity utilizing historical records from the National Hurricane Center’s HURDAT2 archive along with an imagery-centric dataset. A Piecewise Convolutional Neural Network (CNN) act as the foundation for the proposed technique. Satellite images are initially divided into many intensity levels by the framework, after which they are sent to distinct CNN regression models, each of which has been trained for a particular intensity level. Quality of prediction is improved throughout a range of cyclone strengths because to this two-stage system. K-fold cross-validation is used to evaluate the model’s performance. Six-hourly wind speed readings and all recorded hurricanes in the Atlantic and Pacific basins are included in the collection. Metrics like Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE) are incorporated to measure the approach. The model attains an MAE of 7.67 knots and an RMSE of 10.09 knots, showing that it can estimate cyclone intensity with relatively low error. Overall, the approach gives a practical option for intensity monitoring and shows potential for real-time forecasting or future extension to multimodal meteorological inputs.