Fungi image classification is highly challenging due to the high degree of similarity in the visual features and varying image quality. The classification of rare fungi is made more difficult by the limited data and images available. To address this issue, we employ few-shot learning techniques on the Danish Fungi 2020 dataset, utilizing the LibFewShot implementation. In particular, our study focuses on experimenting with five metric-learning based few-shot learning methods and comparing their performances on this dataset. Further, we examine the effectiveness of applying five data augmentations on each method, and find that adding all beneficial augmentations does not yield better results than applying the most beneficial augmentation alone. We also attempt to enhance the models with two self-supervised learning tasks, where we discover them to have the best performance on weaker models. Similarly, when adding augmentations to self-supervised tasks, the overall performance was weakened. Overall, we have found the Cross Attention Network with ColorJitter augmentation to be the optimal model in this application along with a remarkable scalability. Our study provides insights into the potential of utilizing few-shot learning to classify uncommon fungi and directions for further improvements.