In order to facilitate the analysis of the disaster impact caused by hail on social production, it is necessary to make classified statistics on hail magnitude and quantitative analysis on hail magnitude, which can not only provide the basis for disaster assessment, but also give feedback to weather forecast and false report. In this paper, an improved complementary set empirical mode decomposition (CEEMD) reconstruction algorithm is proposed for hail sound signal. The reconstructed signal retains the original time domain characteristics to the greatest extent, and can also denoise hail sound signal. Secondly, a multi-domain feature fusion 1D-CNN model is designed. The reconstructed original data, time-domain features and frequency-domain features are used as the input of 1D-CNN,the features are spliced in the middle layer, and finally the classifier is output. The results show that the recognition rate of the multi-domain feature fusion 1D-CNN designed in this paper is as high as 99.58%, which is 8.75% higher than that of the original data and the traditional 1D-CNN model.