Abstract：In order to explore the process of fully mechanized coal caving mining method for determining coal-rock character recognition, according to differences in physical and mechanical parameters of coal, gangue, roof rock, there will be differences in amplitude, frequency band energy, and frequency for the acoustic pressure signals when the caving collapse impact the end of the beam of hydraulic support. Top coal caving experiments were carried out at underground mine with full-mechanized caving face. A large number of first-hand acoustic pressure signals were obtained under three working conditions (top coal caving, gangue caving, roof rock caving). Wavelet denoising was used to reduce the working noise of scraper conveyor. The signal after noise reduction in three different working conditions was divided by 6 stages wavelet packet, and normalized energy in the 6th stage 64th bands was extracted by frequency band energy analysis method. Then the distribution of energy percentage and statistical index also was analyzed. Through the above-mentioned research, the signal distribution of energy percent of each frequency band and the energy distribution in 0-1 250 Hz, 0-2 500 Hz in three working conditions is various, and they are sensitive to the statistical index in different degree. Research results lay the foundation of improvement of automation and intelligent in fully mechanized top coal caving.