In order to detect the early weak faults of rolling bearings effectively,a method of adaptive stochastic resonance based on variational mode decomposition（ VMD） is proposed. Firstly,the fault vibration signal of rolling bearing VMD is decomposed into finite Intrinsic Mode Functions（ IMF）,IMF components containing fault feature are selected and reconstructed. The reconstructed signal is inputted into the stochastic resonance system,and the system structure parameters are optimized by the quantum particle swarm algorithm with the objective function of the improved weighted kurtosisso as to get the best resonance output,thus realizing the signal de-noising and enhance the fault feature. Finally the fault feature frequency is extracted by the autocorrelation envelope spectrum of output signal. The validity and advantages of the proposed method are proved by the analysis of the measured data.