Deep Learning System for Plant Species Identification and Botanical Knowledge Retrieval

TerraHerb is a Python-first computer vision system for plant species/disease identification with local and remote botanical knowledge enrichment.

Performance Benchmarks

Metric	MobileNetV2 (PyTorch)	EfficientNetB0 (TensorFlow Strategy 98)
Top-1 Accuracy	92.8%	97.8%
Top-5 Accuracy	98.5%	99.2%
Inference Latency	~120ms	~155ms
Dataset	PlantVillage (38 classes, ~54K images)	PlantVillage

Python/ML Architecture

graph TD
    UI[React Web UI] --> API[FastAPI Gateway]
    API --> INF[PlantPredictor]
    INF --> CLS[MobileNetV2 Classifier]
    API --> KR[KnowledgeRetriever]
    KR --> UCI[UCI Plants Local Data]
    KR --> GBIF[GBIF API]
    KR --> WIKI[Wikipedia API]
    TRAIN[train_model.py] --> WEIGHTS[models/saved/*.pth]
    WEIGHTS --> CLS

Loading

Repository Structure

terraherb/
|-- terraherb/
|   |-- api/main.py
|   |-- inference/
|   |-- knowledge/
|   |-- models/
|   |-- datasets/
|   `-- training/
|-- frontend/                  # Vite + React web UI
|-- tests/                     # Python tests
|-- configs/                   # Training configs
|-- datasets_substrate/        # Raw/processed/external datasets
|-- docs/
`-- scripts/

Quick Start

python -m venv venv && source venv/bin/activate
pip install -r requirements.txt
pip install -e .

# Optional: download PlantVillage via KaggleHub
python -m terraherb.scripts.ingest_data

# Train (PyTorch)
python -m terraherb.training.train_model --config configs/default_training.yaml

# Serve API
uvicorn terraherb.api.main:app --host 0.0.0.0 --port 8000 --reload

# Run tests
pytest tests/ -v

Strategy 98 (TensorFlow)

python -m terraherb.training.train_tf

Web UI

cd frontend
npm install
npm run dev

Health Check

./scripts/check-health.sh