AgroVision-Net

A deep learning system that detects and classifies 38 plant disease categories from leaf images. Built with EfficientNetB0 transfer learning and deployed as a Gradio web app.

Upload a photo of a plant leaf → get an instant diagnosis with disease description, severity, and treatment advice.

---

Results

Model	EfficientNetB0 (fine-tuned, two-phase)
Dataset	PlantVillage — 54,305 images
Classes	38 across 14 crop species
Expected accuracy	93–97% on validation set
Input size	224 × 224 RGB
Confidence threshold	50% — low-confidence predictions rejected

Supported crops & diseases

Apple (scab, black rot, cedar rust, healthy) · Blueberry · Cherry (powdery mildew) · Corn (cercospora, common rust, northern leaf blight) · Grape (black rot, esca, leaf blight) · Orange (citrus greening) · Peach (bacterial spot) · Bell Pepper (bacterial spot) · Potato (early blight, late blight) · Raspberry · Soybean · Squash (powdery mildew) · Strawberry (leaf scorch) · Tomato (bacterial spot, early blight, late blight, leaf mold, septoria, spider mites, target spot, yellow leaf curl virus, mosaic virus)

---

Quickstart

1. Clone & install

git clone https://github.com/Sowaiba-01/AgroVision-Net.git
cd AgroVision-Net
pip install -r requirements.txt

2. Download the dataset

Download PlantVillage from Kaggle and extract so the structure is:

data/
└── PlantVillage/
    ├── Apple___Apple_scab/
    ├── Apple___healthy/
    ├── Tomato___Late_blight/
    └── ...

3. Train

# Linux / Mac
export DATA_DIR="data/PlantVillage"
python -m src.train

# Windows
set DATA_DIR=data/PlantVillage
python -m src.train

Training runs in two phases automatically:

• Phase 1 (up to 15 epochs): classification head only, backbone frozen, LR = 1e-3
• Phase 2 (up to 10 epochs): top 20 EfficientNetB0 layers unfrozen, LR = 1e-5

Saves models/best_model.keras and models/class_names.json.

4. Run the web app

python app.py

Open http://localhost:7860 in your browser.

5. Run CLI prediction

python -m src.predict --image path/to/leaf.jpg

---

Running Tests

Tests use mocked models — no GPU, no trained model, no internet required.

# Run all tests
pytest tests/ -v

# With coverage report
pytest tests/ -v --cov=src --cov-report=term-missing

Tech Stack

Model	EfficientNetB0 via TensorFlow/Keras
UI	Gradio
Image processing	Pillow, NumPy
Testing	pytest, pytest-cov
CI	GitHub Actions

---

License

MIT — see LICENSE.