E-Waste Detector

Live Web App:
Try it now — Point your phone camera! Binary classifier (e-waste vs non-e-waste) trained with MobileNetV2 on a balanced dataset of 4,800 images.

Metric	Value
Validation Accuracy	95%
Precision / Recall / F1	0.95 for both classes
Model file	ewaste_detector.h5 (~14 MB)

---

The model file is binary

ewaste_detector.h5 is a saved Keras neural network – do NOT open it in a text editor (you’ll see gibberish).
Treat it like a .zip or .exe file.

Correct way to use it: load with TensorFlow/Keras (Python).

---

How to use the model

# 1. Clone the repo
git clone https://github.com/PrathamB001/e-waste-detector.git
cd e-waste-detector

# 2. Install TensorFlow
pip install tensorflow

# 3. Load and test
from tensorflow.keras.models import load_model
from PIL import Image
import numpy as np

model = load_model("ewaste_detector.h5")
print("Model loaded!")

def predict(img_path):
    img = Image.open(img_path).convert("RGB").resize((224, 224))
    arr = np.array(img) / 255.0
    arr = np.expand_dims(arr, axis=0)
    prob = model.predict(arr, verbose=0)[0][0]
    label = "E-WASTE" if prob < 0.5 else "NON E-WASTE"
    return label, prob

label, prob = predict("your-photo.jpg")
print(f"{label} ({prob:.1%})")

📦 Cloud Data Logging (Firebase Integration)

The app securely logs every detection result to Firebase Firestore, enabling real-time tracking and future analysis of model performance.

---

🔹 Logged Fields

Each detection entry stores the following metadata:

Field	Description
timestamp	UTC ISO timestamp when prediction occurred
label	Model’s classification output (E-WASTE / NON E-WASTE)
confidence	Model confidence score (0–1)
method	Input type used (Camera or Upload)

---

🔹 Firestore Setup

1. Create Firebase Project

• Created a Firebase project.
• Enabled Firestore Database in test mode during development.

2. Generate Service Account Key

• Generated a .json service account key.
• Stored securely using Streamlit Secrets Management.

3. Initialize Firestore in Code

from firebase_admin import credentials, firestore
import json, tempfile

# Load secret key from Streamlit
key_json = json.loads(st.secrets["FIREBASE_KEY"])

# Write key to temporary file
with tempfile.NamedTemporaryFile(delete=False, suffix=".json", mode="w") as f:
    f.write(json.dumps(key_json))
    temp_key_path = f.name

# Initialize Firebase app
cred = credentials.Certificate(temp_key_path)
firebase_admin.initialize_app(cred)
db = firestore.client()

4. Log Each Inference Result from datetime import datetime, timezone

timestamp = datetime.now(timezone.utc).isoformat()

db.collection("detections").add({ "timestamp": timestamp, "label": label, "confidence": float(conf), "method": upload_option })

🔒 Security

Firebase key is never exposed in the repository. Stored securely via Streamlit Secrets Management. Loaded at runtime using st.secrets["FIREBASE_KEY"].

📊 Firestore Dashboard