Learning features that are good for machine learning can be included, especially when dealing with limited data. A difficult case is a one-shot study where the prediction is used for only one example per class. In this work, an attempt is made to delve into the utilization of Siamese neural networks, which possess a distinct architecture for intuitively ranking similarity between inputs. When these networks are finetuned, they provide discrimination capabilities that can be extended to include not only new information but also unknown categories in unknown distributions. It has been observed that an impressive performance can be achieved surpassing other deep learning models and approaching various levels in one-shot classification problems.