Forecasting and predicting coalfield fire and coal concealed fire are difficult components of coal fire control, and there is currently a lack of widely applicable effective methods for these processes. Electromagnetic radiation (EMR) has the advantage of multi-direction cross-locating and has been widely used to monitor and provide early warnings for coal and rock dynamic disasters. In this paper, the time-varying characteristics of EMR during 3 metamorphic grades of the coal-heating process were studied. EMR signals were clearly generated in different metamorphic grades of the coal-heating process. The variation trend of the EMR signals was different for 3 metamorphic grades of coal. According to R/S statistical analysis of the EMR signal time series, all Hurst values were greater than 0.5 and the correlation coefficient was above 0.9. There is a positive correlation between the EMR signals and coal temperature. In the coal-heating process, the EMR signal intensity varied if the EMR test frequency was 1kHz, 15kHz and 50kHz. Thermal deformation and thermal cracking of the coal body occurred in the coal-heating process. The free electrons and electric dipole in coal migrate and generate EMR. Based on the changing characteristics of the EMR in the coal-heating process, we present the conceptual design of a method to detect concealed fire in underground mines.