Smart Waste Classification System (FastAPI + ESP32)

This project implements a real-time smart waste classification system using FastAPI, ESP32-CAM, and ESP32 microcontrollers.
The ESP32-CAM captures waste images and sends them to this FastAPI server via WebSocket.
The server uses a pre-trained TensorFlow/Keras model to classify the image as non-biodegradable (0) or biodegradable (1), and sends the classification back to the ESP32 for actuation (e.g., servo bin sorting).

---

Features

• FastAPI backend with WebSocket support.
• Real-time communication between ESP32-CAM and ESP32.
• TensorFlow/Keras model for waste image classification.
• Automatic image saving and classification feedback.
• Health check and image retrieval HTTP endpoints.
• Modular design — can integrate with IoT dashboards or cloud services.

---

System Architecture

ESP32-CAM → Captures image → sends to → FastAPI Server (this code) → classifies image → sends result to → ESP32 (servo/sensor)
The ESP32 controls sorting mechanisms or provides visual feedback.

---

Requirements

Python dependencies

• Ensure you are in the waste-AI directory

  • Install required packages:

    pip install -r requirements.txt

  • Create a virtual environment and activate it

    python -m venv env
    

  • Activate the virtual environment by

    env\Scripts\activate
    

  • To start the server

    python -m waste
    

Endpoints

HTTP Endpoints

Method	Endpoint	Description
GET	/health	Checks if the server is running.
GET	/latest	Returns the latest image captured by the ESP32-CAM.

WebSocket Endpoints

Endpoint	Connected Device	Description
/ws/image	ESP32-CAM	Receives binary image data and sends classification result.
/ws/sensor	ESP32	Receives sensor data (e.g., object detected) and triggers ESP32-CAM capture.

---

Data Flow

1. ESP32 detects an object and sends "object_detected" → /ws/sensor.
2. Server tells ESP32-CAM to "capture" → /ws/image.
3. ESP32-CAM sends the image as binary data → /ws/image.
4. Server saves the image (latest.jpg) and classifies it using tf_model.h5.
5. Server sends classification result ("0" or "1") → /ws/sensor.
6. ESP32 actuates servo or performs next action.
7. Server instructs ESP32 to "resume_sensing" after processing.

Example Console Logs

INFO: ESP32-CAM (image) connected via WebSocket.
INFO: Received image data from /ws/image: 24567 bytes
INFO: Latest image saved to latest.jpg
INFO: Classified as: Non-biodegradable (1)
INFO: Sending classification '1' to Esp32 (/ws/sensor).
INFO: Sent 'resume_sensing' signal to Esp32 (/ws/sensor).

---

Model Integration

• The ML model file tf_model.h5 should be placed inside the /model directory.
• Input shape: (150, 150, 3)
• Output: A single sigmoid neuron → value <0.5 = biodegradable (0), >=0.5 = non-biodegradable (1)

---

Notes

• If the model fails to load, the server will return a neutral value (0.5) for predictions.
• Images are resized to 150x150 pixels and normalized to [0, 1].
• Temporary image files are automatically deleted after prediction.

---

Viewing Latest Captured Image

After at least one capture, open your browser at:

http://<server-ip>:8000/latest

You’ll see the most recent image sent by the ESP32-CAM.

---

🧑‍💻 Author

Developed by: [Theophilus Adeleke]

Tech Stack: FastAPI · TensorFlow · Keras · ESP32 · WebSockets · Python

---

📜 License

This project is licensed under the MIT License — feel free to use, modify, and distribute.

---