Abstract: Safety is essential to deep mining operations. To monitor and detect the safety condition of surrounding rock of roadway and deep coal mining site, many methods obtain competitive results by monitoring targets situation based on laser scanning spacial information. The Hyperspectral LiDAR (HSL) technology can acquire spacial and spectral data for deep mine safety detection and further fine structure analysis. The exact coal/rock classification is the basis of detection and analysis. While, in on-site operation, HSL signals are susceptible to instrument attributes and environmental factors, and need calibration for further classification application. However, due to serious dust pollution in deep coal mines, conventional calibrations are hard to achieve the desired results. To address this issue, a new method was proposed to classify coal/rock without calibration. First, the new feature values, waveform entropy (WE) and joint skewness-kurtosis figure (JSKF), were extracted from coal/rock samples based on HSL measurements. Then, the coal/rock classification tests were conducted with random forest (RF) and support vector machine (SVM) classifiers. Additionally, the spectral properties of different wavebands were evaluated by spectral segmentation test and the classification performances were optimized further by selecting specific channels. The results show that the proposed method can achieve excellent classification accuracy for coal/rock without calibration.