Abstract: Accurate and rapid identification of coal types provides important technical support for ensuring the efficient operation of the coal industry. To address the problems of long detection cycles, complex operational procedures, and high equipment costs associated with traditional coal classification methods, a coal type classification method based on the transmission characteristics of multi-frequency electromagnetic waves is proposed. A USRP B210 software-defined radio platform and the GNU Radio open-source software framework are employed to construct a signal generation and acquisition system in a virtual machine environment. Six coal types—lignite, bituminous coal, anthracite, coking coal, gas coal, and lean coal—are selected as the research objects. Within the frequency band from 3 GHz to 6 GHz, transmission signals at 301 frequency points are collected with a frequency step of 10 MHz. By setting four different single-frequency-point scanning durations of 0.1 s, 0.2 s, 0.4 s, and 0.8 s, the transmission signals are acquired and the energy attenuation value at each frequency point is calculated. A hybrid architecture coupling a convolutional neural network with a long short-term memory network（CNN-LSTM） is adopted to simultaneously extract local frequency-domain features and long-range temporal dependencies of the signals, and a particle swarm optimization（PSO） algorithm is introduced to achieve automatic optimization of the model’s key hyperparameters. Experimental results show that, with the original feature structure unchanged and the coal type identification scope expanded, the classification accuracy is improved by 5.12% compared with the original model, and the average accuracy reaches 97.80% when the single-frequency-point scanning duration is 0.80 s. The coal type identification method based on microwave transmission characteristics exhibits high identification accuracy and fast response speed, and thus possesses certain application value in the field of coal classification.