Abstract: For the damage on-line monitoring of carbon fiber reinforced plastics (CFRP) wrapped gas cylinder, the acoustic emission detection of CFRP cylinders during impact damage was studied. The acoustic emission signal of gas cylinder damage obtained was taken as the research object, and the original signal was converted into the eigencoefficient vector by the method of Mel-Frequency cepstral coefficients feature extraction, and the parameter values and variation trends were compared synchronizarily. The results showed that the distribution of MFCC of different damage types presented obvious regular distribution. This research result can provide some reference for the recognition of acoustic emission detection signals in CFPR materials.