Abstract: In shaft mining, conventional vibration monitoring of blasting operations is often affected by environmental factors and system limitations, complicating the extraction of coal–rock rupture signals. This paper explores the use of electromagnetic radiation signals for blasting monitoring, introducing a noise reduction method for these signals and examining the time–frequency characteristics of the pure signals. Initially, the paper suggests using the dung beetle optimizer (DBO) algorithm to dynamically adjust the parameters of variational mode decomposition (VMD) for the efficient acquisition of optimal decomposition parameters k, α. By analyzing electromagnetic signal optimization sunder different fitness functions and evaluating three types of anomalies, namely repeated mutations, boundary stabilization, and unchanged states, we find that the performance of the DBO-VMD model in processing coal–rock electromagnetic signals ranks as follows: envelope entropy > ranking entropy > information entropy > sample entropy. Α center-frequency criterion noise reduction model is proposed to eliminate high-, intermediate-, and low-intensity components in the signal. When comparing electromagnetic signals processed by the DBO-VMD and empirical mode decomposition (EMD), the DBO-VMD effectively avoids modal aliasing and provides more reasonable center-frequency distributions. After applying a consistent noise reduction process, the DBO-VMD model shows superior performance over EMD. It provides enhanced smoothing and fidelity of pure signals and is more efficient at noise screening. The DBO-VMD achieves a signal-to-noise ratio about two times that of the EMD. Finally, we conducted a statistical analysis of the entropy, energy, bispectrum, and time–frequency domain characteristics of pure electromagnetic signals associated with coal–rock ruptures. During stable periods, information entropy, instantaneous energy, and marginal energy remain below specific thresholds, but they exhibit sudden changes during rupture events. Rupture periods begin when information entropy falls below 4.75, instantaneous energy exceeds 1000 J, or marginal energy surpasses 100 J, based on a 50-point time window. Conversely, the conclusion of the rupture period corresponds to opposite conditions, with marginal energy responding more sensitively to rupture states than instantaneous energy. During ruptures, skewness is positive, steepness ranges from 0.9 to 4.6, and pulse index varies from 3.7 to 6.1, all within a frequency band below 20 kHz. Main rupture events coincide with peak signal energy, mostly under 5 kHz. As frequency increases, signal amplitude decreases rapidly, with low-energy pulses during non-rupture periods concentrated in the 0–3 kHz range. This study sheds light on the time–frequency characteristics of electromagnetic radiation signals generated by blasting. These insights lay the groundwater for effectively monitoring such signals in underground mining operations.