Aiming to address issues such as insensitivity to fault characteristics, inadequate feature extraction, and susceptibility to external environmental interference in traditional bearing fault diagnosis methods, an improved approach to rolling bearing fault diagnosis based on ConvNext V2 is proposed. Firstly, leveraging the advantages of wavelet time-frequency analysis, one-dimensional vibration signals collected from the experimental setup are transformed into two-dimensional images containing abundant time-frequency information using continuous wavelet transform. Secondly, the next-generation convolutional neural network ConvNext V2 is introduced as the underlying network for rolling bearing fault diagnosis, and the Efficient Channel Attention (ECA) network is incorporated to enhance the feature extraction capability of the Block module in the base network. Subsequently, the wavelet time-frequency images are used as feature map inputs to the improved ConvNext V2 network for training, enabling the diagnosis and recognition of rolling bearing faults. Finally, experimental results demonstrate that the proposed method exhibits good fault recognition accuracy and stability in laboratory data, with high generalization capability, thereby offering a new solution for rolling bearing fault diagnosis.