To improve the accuracy of fault diagnosis of shaft bearing of mine hoist under strong noise influence, this paper proposes a method combining variational mode decomposition （VMD）, multipoint optimal minimum entropy deconvolution adjusted （MOMEDA）, and convolutional neural network (CNN). The Sparrow search algorithm combining sine-cosine and Cauchy mutation is used to perform multi-objective optimization of the penalty factor and decomposition levels of VMD. The vibration signal is decomposed by VMD according to the kurtosis criterion to obtain intrinsic mode functions (IMF). The intrinsic mode functions containing shock components are selected to reconstruct the original signal. MOMEDA is applied to the reconstructed signal for noise reduction. An autocorrelated kurtosis index is established as the fitness function to optimize the key parameter, fault period T, of MOMEDA; permutation entropy is used as the objective function to optimize the filter length. The signal enhanced by MOMEDA is envelope-demodulated, and the envelope amplitude sequence is used as a feature input to the CNN model for training and validation to obtain fault diagnosis results. The methods of VMD-MED-CNN, VMD-MCKD-CNN and VMD-CNN are compared and analyzed. The results show that the average accuracy of VMD-MOMEDA-CNN proposed in this paper is the highest, reaching more than 98%. It is proved that the algorithm has superior accuracy and stability under the influence of strong background noise.