Based on the uniaxial compression experiment of granite and decision tree model

the extraction method of key acoustic emission signals in rock fracture process is constructed. The extracted key signals are firstly classified by choosing the characteristic parameters

and the time-frequency features of the various key acoustic emission signals are analyzed

then the rock fracture mechanisms corresponding to these signals are discussed. The results show that with a signal recognition accuracy of over 90%

the method based on decision tree model can effectively extract the acoustic emission signals corresponding to the critical fracture events

which affect the stability of the whole rock structure during the rupture process. The key signals are divided into four categories according to the feature extraction rules. The signals of Class A correspond to the macroscopic cracking and expanding fracture of rock. The signals of Class B correspond to a large number of small-scale fractures and extensional cracks that occurred in the near and post-peak stages of the fracture process. The signals of Class C correspond to the shear-slip fracture before the rock experiencing macro-fracture. The signals of Class D correspond to the small-scale tension-shear composite fracture after the whole rock failed.