<![CDATA[Indonesia has a high biodiversity of fungi, including edible and toxic species. Manual identification is often challenging due to morphological similarities between safe and poisonous species. Therefore, this study evaluates the use of deep learning-based Convolutional Neural Network (CNN) with the MobileNetV2 architecture for mushroom classification. The research method includes collecting a dataset of 1,500 images from 10 mushroom species (5 edible and 5 toxic), preprocessing data by normalizing image size and applying augmentation techniques, and training the model using the Adam optimizer with dropout and early stopping to prevent overfitting. Hyperparameter tuning was performed using grid search on batch size (64, 128, 256), epochs (20, 50, 100), and learning rate (0.1, 0.01, 0.001). The test results show that a combination of batch size 64, epoch 50, and learning rate 0.1 achieved 98% validation accuracy. The final model was tested and achieved 95.33% accuracy, with an average precision, recall, and f1-score of 95%. These results confirm that MobileNetV2 is effective in classifying mushroom species and can assist in more accurately identifying edible and toxic fungi.]]>
Copyrights © 2025
