Fine-Grained Classification of Edible Mushrooms Using Convolutional Neural Network

Previous research mainly focused on categorizing mushrooms based on their edibility using varying features and non-image data types. Thus, fine-grained classification using deep learning approaches remaius understudied. The present study introduced a fine-grained mushroom classifier that uses a convolutional neural network (CNN). The system identified eight edible mushroom species with a created mushroom dataset comprised of 9,000 images with the developed classification system using transfer learning and k-fold cross-ralidation techniques. Transfer learning involved using pre-eristing CNN modek, such as ResNetRS50 and EfficientNetV2B0 trained on ImageNet, as feature extractors. The k-fold cross-ralidation algorithm determined the optimal candidate model between the two for the mushroom dataset. EfficientNetV2B0 emerged as the most effective model, with a validation accuracy of $90 \%$, thus serving as the basis for the final mushroom classification system. The final classification system demonstrated an average testing accuracy of $92 \%$, proving its ability to identify the most edible mushrooms in the dataset accurately.