The fault impact signal of bearing on large-scale wind turbine is usually disturbed by complex loads and strong background noise,and its incipient faults are not easy to detect. To solve this problem,in this paper,an improved multi-scale morphological analysis method based on information entropy（ IE） and feature energy factor（ FEF） is proposed. Morphology gradient product operation（ MGPO） is an effective morphology operator for extracting rolling bearing impact signals. This paper proposes a multi-scale morphological analysis method based on MGPO operator in order to extract more detailed fault feature information,and for improving the inadequacy of the kurtosis criterion and the signal-to-noise ratio in selecting the optimal scale,in this paper,a comprehensive scale range selection method based on information entropy and characteristic energy factor is also proposed. Experimental and comparative results show that the algorithm proposed in this paper has a certain of advantages.