<p>Accurately identifying rare rice varieties presents a significant challenge due to the inherent scarcity of labeled data, rendering traditional deep learning approaches infeasible. While few-shot learning offers a promising alternative, existing methodologies often struggle to effectively capture the subtle visual distinctions crucial for differentiating between closely related rare categories with limited examples. To address this, we introduce Fs-SF2E, a novel few-shot learning framework for rare rice recognition that synergistically integrates strategic meta-learning foundations with a dedicated feature enhancement module. Our approach employs an episodic training paradigm guided by a standard uniform task sampling strategy, enabling robust learning from minimal data. Furthermore, it incorporates a combination of targeted data augmentation techniques and a discriminative feature embedding module designed to amplify subtle yet critical visual cues unique to each rare rice variety. We rigorously evaluate the efficacy of Fs-SF2E on two challenging datasets: miniImageNet and T-Rice, a specialized benchmark for diverse Vietnam rice varieties. Extensive experimental results demonstrate that our framework significantly outperforms state-of-the-art few-shot learning methods, achieving remarkable recognition accuracy for rare rice categories and highlighting its potential for real-world applications in agriculture and food authentication.</p>