Aiming at the problem that it is difficult for the early weak feature of the bearing to separate from the strong noise background

a method of extracting the feature based on the intrinsic time-scale decomposition（ ITD）and adaptive generalized morphological filtering is proposed. ITD is decomposed into several proper rotation（ PR）components

and several PR components are selected by using the kurtosis criterion. The reconstructed signal is processed by adaptive generalized morphological filtering. The signal after noise reduction is reconstructed by Hilbert envelope analysis. Through the experimental analysis of the inner and outer rings of the rolling bearing

the validity of the method is verified.