Extraction of repetitive shock signals from vibration signals is crucial in bearing fault detection. Morphological filtering can extract the signal components based on the geometric features of the signal. In this paper, we propose an improved morphological filter for extracting fault cycle pulses from vibration signals in the time domain. Morlet wavelets are used as the main waveforms of the structural elements, which guide the construction of the structural elements based on their correlation with the defect-induced pulse, the Teager energy operator (TEO) is used to reduce the modulated signal interference. The improved morphological filtering algorithm can obtain fault features from low signal-to-noise ratio signals. The processing results of one simulated signal and two sets of experimental signals verify the effectiveness and robustness of the method.