Along with the development of modern industry, it is increasingly important to diagnose and detect faults for timely maintenance of equipment. Intelligent fault diagnosis for rotating machinery covers a wide range of research subjects, including motors, gearboxes, turbines, and many others. Bearings play an extremely important role, as they cause most of the faults in rotating machinery. Deep learning has shown promise in bearing failure analysis; however, there are still many challenges due to limited labeled data. Few-shot learning approaches address this problem, but traditional methods rely heavily on CNN-based feature extraction. In this chapter, we propose a novel framework integrating Mamba, and selective aggregation feature extractor enhanced by priority channel and spatial attention, that helps improve long-range dependency modeling and feature selection. The extracted features are processed through two complementary branches: a transformer with cross-attention for contextual learning and a Mahalanobis distance-based module for metric-based classification. We evaluated our method on two datasets, PU and HUST bearing. The experimental results show that our model outperforms state-of-the-art methods, highlighting its effectiveness in the task of classifying small-sample bearing faults.