In view of the strong noise environment traditional bearing fault diagnosis method for fault recognition rate low, is put forward based on the discrete wavelet transform (discrete ‘wavelet transform, DWT and the improved deep residual shrinkage network (IDRSN) fault diagnosis model. Firstly, DWT is used to convert the bearing vibration signal into a two-dimensional time-frequency graph. Secondly, an improved soft threshold function (ISTF) was designed. Using improved soft threshold block (ISTB) and adaptive slope block (ASB), An improved residual shrinkage building unit (IRSBU) was constructed, and the proposed method was verified experimentally with Case Western Reserve University rolling bearing data set. The results show that the proposed fault diagnosis method has better classification accuracy than other methods.