In the long-term use process, the performance of rolling bearing will be degraded to different degrees. If the degradation state of rolling bearing can be identified online, accidents can be effectively prevented. In this paper, an adaptive noise-assisted collective empirical mode decomposition (CEEMDAN) method combined with energy entropy is proposed to extract the characteristics of vibration signals, and then the characteristics are input into the DSHDD model, and the obtained results are input into the membership function to calculate the membership, which can be used as the evaluation index of performance degradation. An adaptive threshold is set using 3 sigma to determine the bearing"s early failure threshold. CEEMDAN and Hilbert envelope demodulation methods were used to verify the correctness of the evaluation results. Finally, the validity and practicability of the model are verified by using the bearing life cycle data from the university of Cincinnati.