Early fault signature detection and background noise removal are essential for bearing fault diagnosis. A novel multiscale enhanced morphological top-hat filter fault diagnosis method, adaptive variational mode decomposition-sample entropy-multiscale enhanced top-hat filter (AVMD-SE-MEMTF), is proposed based on AVMD-SE noise reduction. First, gray wolf optimization algorithm is proposed to optimize the VMD to achieve the optimal decomposition parameters adaptively and combine with SE to eliminate the high noise components and improve the noise reduction effect. Then, based on the pulse extraction property of morphological operations, the concept of MEMTF is proposed. To enhance the multiscale index selection strategy, a synthesis method of eigenfrequency envelope coefficients is constructed to increase the accuracy of the operator during the vibration signal process. Finally, experimental and engineering results show that the proposed method has good diagnostic performance for weak faults in the presence of noise interference.