<![CDATA[This reserach proposes a mushroom species classification method based on digital image processing using a Convolutional Neural Network (CNN). The EfficientNet-B4 architecture was adopted as the backbone model, employing a transfer learning approach followed by a fine-tuning process. The dataset consisted of 3,000 mushroom images categorized into 10 classes, with each class containing 300 images. The model implementation was carried out using Google Colab and the Python programming language. Model performance was evaluated using accuracy, precision, recall, and F1-Score metrics. Several model variations were examined by adjusting training parameters and data split ratios. The best-performing model, referred to as Model 1, utilized a customized freeze layer and applied an 80% training, 10% validation, and 10% testing data split, achieving the highest performance with 90.00% accuracy, 90.09% precision, 89.63% recall, and an 89.59% F1-Score. The findings indicate that applying a customized freeze layer effectively reduces the number of trainable parameters, leading to improved model accuracy. Furthermore, the selection of data split ratios contributes to performance differences during the training and testing phases.]]>
Copyrights © 2026
