7Dagm21/Crop-yield-prediction-and-disease-detection

This project materializes an Ethiopian Crop Intelligence Dashboard built using Streamlit. # Ethiopian Crop Intelligence System A PyTorch-based agriculture intelligence system for Ethiopian grains and coffee. ## Scope This project is limited to these Ethiopian crops: - Teff - Wheat - Maize - Sorghum - Barley - Millet - Chickpea - Coffee The system has two main modules: - Crop yield prediction A concise guide to run and develop the Ethiopian Crop Intelligence System (yield + disease). Prerequisites - Python 3.10+ and a virtual environment - Install project dependencies: ```powershell python -m venv .venv .venv\Scripts\pip install -r requirements.txt ``` Quick start - Start the Streamlit dashboard (local dev): ```powershell streamlit run main.py ``` Public deployment - If you want the dashboard visible from any computer, deploy it to Streamlit Community Cloud and set the entry file to `main.py`. - See `deployment/streamlit-cloud.md` for the exact steps and the repo size warning. - Run the FastAPI server (inference): ```powershell uvicorn src.api.app:app --host 0.0.0.0 --port 8000 ``` Run tests - Execute unit tests: ```powershell pytest -q ``` Training (disease models) - Expect ImageFolder layout under `data/disease_images/` with `train/` and `test/` subfolders per class. - Train a disease model (example): ```powershell python -m src.training.train_disease_model --data data/disease_images --crop coffee --epochs 5 ``` Model export - Export TorchScript/ONNX for a saved bundle: ```powershell python scripts/export_model.py --bundle models/disease/coffee_disease_model.pth --out models/exports/ ``` Repository notes - Large training datasets and model checkpoints live in the workspace but should not be pushed to public remotes unless intended. Keep them in `data/` and `models/` locally. - Common cache folders to ignore: `__pycache__`, `.pytest_cache`, `.ipynb_checkpoints`. Project structure (key folders) ``` ethiopian-crop-intelligence-system/ ├─ data/ # datasets (local only) ├─ models/ # saved mo …