EcoCompress

EcoCompress is an open-source, low-power AI/ML-based data compression toolkit designed for resource-constrained edge devices and IoT sensors. It enables efficient, adaptive, and high-fidelity compression of time-series sensor data using lightweight neural networks, with a deployment-ready workflow for STM32 microcontrollers.

---

🚀 Features

• Lightweight 1D Convolutional Autoencoder for time-series sensor data
• High compression ratios with tunable fidelity (MSE/PSNR metrics)
• Preprocessing pipeline: normalization, windowing, train/val/test split
• Model optimization: quantization, ONNX export, TensorFlow conversion
• Deployment-ready: STM32Cube.AI and STM32CubeIDE integration for STM32 MCUs
• Clear, modular Python scripts for every step

---

📂 Project Structure

EcoCompress/ ├── data/ # Raw and processed datasets (.csv, .npy) ├── src/ # All Python scripts │ ├── preprocess_window.py │ ├── train_autoencoder_tuned.py │ ├── evaluate_autoencoder.py │ ├── optimize_model.py │ ├── onnx_to_tf.py ├── notebooks/ # (Optional) Jupyter notebooks for EDA/prototyping ├── docs/ # Documentation ├── firmware/ # STM32 deployment files and code ├── tests/ # Unit tests (optional) ├── requirements.txt # Python dependencies ├── README.md

---

🛠️ Quick Start

1. Clone the Repository

git clone https://github.com/Paramveersingh-S/compressor.git cd compressor

2. Install Dependencies

python -m pip install -r requirements.txt

3. Preprocess Data

Place your sensor data CSV (e.g., AirQualityUCI.csv) in data/, then run: python src/preprocess_window.py

4. Train the Model

python src/train_autoencoder_tuned.py

5. Evaluate Model Fidelity

python src/evaluate_autoencoder.py

6. Optimize and Export Model

python src/optimize_model.py

7. Convert ONNX to TensorFlow (for STM32Cube.AI)

python src/onnx_to_tf.py

8. Deploy to STM32

• Use STM32Cube.AI to import the ONNX or TensorFlow model.
• Generate and integrate C code into your STM32CubeIDE project.
• Flash and run on your STM32 Discovery board!

---

📊 Results & Benchmarks

• Compression Ratio: Achieves up to 10:1 (lossy, tunable)
• Fidelity: PSNR > 21 dB (can be increased with tuning)
• Model Size: < 50 KB after quantization (STM32-ready)
• Supported Hardware: STM32 Discovery and Nucleo boards

---

📖 Documentation

• STM32Cube.AI User Guide
• ONNX to TensorFlow Conversion
• Project Wiki (coming soon)

---

🤝 Contributing

Contributions, bug reports, and feature requests are welcome!

• Fork the repo and submit a pull request.
• Open an issue for questions or suggestions.

---

📄 License

This project is licensed under the MIT License. See LICENSE for details.

---

🙏 Acknowledgements

• UCI Air Quality Dataset
• PyTorch, ONNX, STM32Cube.AI
• Inspired by CERN’s Baler project and TinyML best practices.

---