Abstract: Microseismic signal is a typical timevarying nonstationary signal,while the collected data are often disturbed by strong noise,seriously affecting the recognition efficiency of microseismic events and the accuracy of source location. Existing traditional recognition methods still have problems with a low rate,high delay and low accuracy. Therefore,based on wavelet theory,the wavelet scattering decomposition transformation method is proposed to analyze and extract the features of microseismic events and noise events signals,calculate the feature coefficients of the two signals,and form the corresponding feature matrix. Based on the characteristics of lowbandwidth,unpredictability and burst transient of microseismic signals,the effects of timeinvariant scale,the number of scattering decomposition and quality factors on the dimension of the feature matrix are compared and analyzed by processing the real data. By adjusting the size of the parameters,the dimension of the feature matrix is effectively controlled to improve the operation efficiency. Finally,the optimal feature matrix is selected for the training of the machine learning classification model. The experimental results show that the feature matrix composed of the wavelet scattering coefficients can effectively distinguish the event signals from the noise signals,and the feature coefficients change trend of the event signals is mutant,and the noise signals feature coefficients change trend appears as a disorder. The feature matrix composed of three times wavelet scattering decomposition transformation can effectively represent the different characteristics of microseismic event signals and noise signals and improve programming efficiency,and the lowfrequency part of event signal is more obvious with the number of transformations increasing. In the case where the timeinvariant scale and the number of transformations are determined,in order to obtain the optimal feature matrix,the quality factor selection of the wavelet scattering transformation of each order should not be too large.