Recognition of Mushroom Species Using Few-Shot Learning Method with a Siamese Network

Authors : Göktürk Öztürk, Köksal Erentürk

Pages : 262-266

View : 91 | Download : 264

Publication Date : 2025-11-30

Article Type : Research Paper

Abstract :Mushrooms, as nutritionally and medicinally valuable macrofungi, require accurate recognition due to the presence of toxic species causing severe health risks. Traditional methods based on morphology are time-consuming and prone to human error, which makes automated solutions essential. In this study, a siamese neural network with a ResNet18 backbone was applied to mushroom species recognition under a 7-way 3-shot learning setting. The dataset, derived from Kaggle, was pre-processed with background removal, resizing, normalization, and augmentation to ensure reliable feature extraction. The model was trained with cosine embedding loss and evaluated using accuracy, precision, recall, F1-score, and confusion matrix analysis. Results demonstrated a high classification accuracy of 90.48%, showing that the model effectively distinguishes mushroom species despite a small number of confusions. These findings confirm the effectiveness of siamese networks for mushroom classification and suggest future improvements.
Keywords : few-shot öğrenme, siamese ağı, mantar tanıma, görüntü sınıflandırma, derin öğrenme

ORIGINAL ARTICLE URL