The accurate monitoring and forecast of tropical cyclone intensity can effectively reduce the cost of disaster preparedness and risk mitigation. In this paper we estimated the real-time intensity of tropical cyclones and predicted cyclone intensity in 6-12 hours through a combination of the Communication, Ocean and Meteorological Satellite (COMS) Meteorological Imager (MI) satellite images and European Centre for Medium-Range Weather Forecasts Era-interim (ECMWF ERA-Interim) numerical model output with deep Learning. This study used tropical cyclones that occurred in the western North Pacific between 2011 and 2016. Two schemes considering different combinations of input data and machine learning methods were evaluated. Scheme 2, which applied the fusion network model by using satellite data and the numerical model output, yielded the higher accuracy than Scheme 1.