Megaplant
A consolidated leaf-image dataset designed to support plant disease classification models that generalize across diverse environmental conditions, from controlled laboratory settings to highly variable in-field scenarios. MegaPlant integrates multiple publicly available datasets and standardizes them into a unified taxonomy of healthy and diseased leaf categories, enabling robust training across modalities.
Methodology
- Aggregated datasets from Kaggle and official repositories, totaling nearly 60,000 plant images
- Used PyTorch in building a convolutional network
- Focuses only on disease detection and symptom identification
Metrics
- Average increase of 8% in classification accuracy relative to foundational datasets
- 95.12% accuracy in classifying plant disease