<p>Data-driven methods have achieved remarkable results in the field of bearing fault diagnosis. However, the variable-speed conditions and strong noise interference severely limit further improvement in diagnostic performance. To address this issue, this paper proposes an intelligent fault diagnosis framework based on feature enhancement. The method first leverages the angle-time cyclostationary characteristics of bearing vibration signals to construct an order-frequency spectral correlation (OFSC) feature map, which suppresses speed fluctuation interference and enhances fault features. Subsequently, a hybrid Vision Transformer-CNN network architecture is designed, where the Transformer module captures global long-range dependency features while CNN extracts local detailed features, enabling collaborative learning of multi-scale features. Finally, the performance of the proposed method is validated through comparative experiments on datasets. The results demonstrate that the proposed method exhibits outperforms under both variable-speed and strong noise conditions, and the diagnostic accuracy based on OFSC features significantly outperforms other feature maps.</p>