As a new type of meteorological detection equipment, millimeter wave cloud radar has certain advantages in detecting cloud particles. This paper uses millimeter wave radar to detect an icing process in Yongzhou, Hunan Province from December 26 to 28, 2021. On the basis of the basic data observed by the millimeter wave cloud radar, the attenuation correction of the echo is carried out by using the convolutional neural network algorithm. Based on the corrected echo reflectivity, this paper has carried out a detailed analysis of different precipitation phases, and found that fog, snow, freezing rain and other phases appeared repeatedly in this icing process, while the growth of line icing mainly occurred during the freezing rain period. The detection results show that the millimeter wave cloud radar can better observe the precipitation particles in the process of non-uniform precipitation in winter, and provide decision-making basis for early prediction and early warning of grid icing.