This paper proposes a new method for predicting the remaining life of rolling bearings, which improves the prediction accuracy by enhancing the characteristics of vibration signals. First, the original vibration signal is extracted using time-domain statistical features to capture the basic degradation characteristics of the bearing. Secondly, a network model based on frequency-domain interpolation and time-domain reconstruction (FITR) is proposed. This model achieves accurate simulation and generation of vibration signals through technologies such as reversible instance normalization, real-valued fast Fourier transform, and complex-valued linear layers. By combining the original signal features with the signal features generated by FITR, a more comprehensive feature representation system is constructed. Finally, the enhanced features are input into the Bi-LSTM network for remaining life prediction. Experimental results on a public full-life dataset show that the method proposed in this paper can effectively improve the prediction accuracy of the remaining life of bearings. This method provides a new solution for health monitoring and life prediction of rolling bearings.