CropcareAI is a high-performance deep learning platform designed for real-time agricultural disease detection and treatment. Built for research excellence and practical field utility, it leverages state-of-the-art computer vision to assist farmers in identifying 38+ crop-disease pairs with professional accuracy.
- Deep Architecture: Developed using EfficientNetV2-B0, utilizing transfer learning for superior feature extraction and parameter efficiency.
- Large-Scale Data: Trained and validated on the full 55,448-image PlantVillage dataset, covering 14 crop species and 42 specific health conditions.
- Explainable AI Prep: Infrastructure configured for advanced metrics including ROC-AUC Curves, Confusion Matrices, and Per-Class F1-Scores.
- Edge Deployment: Optimized for low-latency inference, bridging the gap between high-end deep learning and real-world mobile accessibility.
- Backend: FastAPI (Python 3.12)
- DL Framework: TensorFlow 2.15+ / Keras 3.0
- Database/Auth: JWT-based secure authentication
- Image Processing: PIL, NumPy, OpenCV
- Visualization: Matplotlib, Seaborn, TensorBoard
backend/: Core logic, training pipelines, and REST API.saved_models/: Production-ready model weights and training history.scripts/: Data acquisition and preprocessing utilities.
research/: Scientific artifacts and IEEE paper documentation.plots/: High-resolution performance visualizations.
frontend/: (Optional) React/Next.js dashboard for the diagnostic engine.
- Training Accuracy: ~98.2%
- Validation Accuracy: ~96.5%
- Inference Latency: <300ms per image
- Automatic translation of symptoms and curative measures into Hindi and Marathi, ensuring accessibility for small-scale farmers in various regions.
Developed as part of an IEEE Research Submission for Advanced Agricultural Engineering.