Abstract:
This project focuses on developing a deep learning-based system for the classification
of mushrooms into six categories: Blue Oyster, Oyster, Phoenix Oyster, Pink Oyster,
Poisonous, and Agaricus. A dataset of 2,134 images was prepared, combining 1,500
manually collected images and 634 sourced online. Advanced preprocessing
techniques such as resizing, normalization, and augmentation were applied to
enhance data quality. Several pre-trained models, including VGG16, MobileNetV2,
ResNet50, and InceptionV3, were evaluated, with InceptionV3 achieving the highest
accuracy of 98% after fine-tuning. The system was deployed using a Streamlet-based
web interface, enabling real-time predictions with minimal latency. Evaluation
metrics, including precision, recall, and F1-score, validated the model’s performance.
The project highlights the practical application of deep learning in mushroom
classification and offers a scalable, efficient solution with potential for further
enhancements in transparency and dataset expansion
