Microwave (MW) radiometers carried by low-orbit satellites can penetrate cloud tops and reveal convection and precipitation. However, due to the orbit height and scanning width, it is currently difficult for the MW radiometer to observe the moving tropical cyclone (TC) continuously. On the contrary, Infrared (IR) radiometers carried by stationary satellites have the advantages of high spatio-temporal resolution and wide spatial coverage, but they can only get cloud-top information about TCs. Combining the advantages of the above two types of radiometers, DeepTCTransfer, a generative deep learning model for the transition from IR to MW is developed to apply in TC studies. Experimental results reveal that DeepTCTransfer based on the diffusion model shows good performance on a variety of image and meteorological metrics. Concentric eyewalls obscured by clouds and extreme values difficult to generate by conventional convolutional networks can also be reconstructed with DeepTCTransfer. Attention maps demonstrate how accurately DeepTCTransfer captures the physical properties. The fine-tuning experiment to generate TC surface wind over the ocean demonstrates the transferability of DeepTCTransfer. This method can reconstruct the MW data with the same spatio-temporal resolution as geostationary satellite IR radiometers by using multi-channel IR, and it can make up for the lack of MW data for TCs, and has the potential to provide support for real-time TC monitoring, and improvement of TC intensity and size estimation and prediction.