sreenathsree1578/mobile-phone-usage-detector

Mobile Phone Usage Detection Model

Model Description

This is a deep learning model trained to detect mobile phone usage in images. The model uses Transfer Learning with MobileNetV2 as the base architecture and custom classification layers.

• Model type: Image Classification
• Task: Binary Classification (Using Phone vs No Phone)
• Architecture: MobileNetV2 with custom classifier
• Input shape: (224, 224)
• Output: Binary probability

Intended Uses & Limitations

Intended Use Cases

• Surveillance systems
• Driver safety monitoring
• Classroom monitoring
• Workplace productivity analysis

Limitations

• Performance may vary with different lighting conditions
• May have reduced accuracy with low-resolution images
• Trained on specific dataset - may need fine-tuning for other domains

Training Data

The model was trained on the Mobile Phone Usage Dataset from IITR

• Training samples: 711
• Validation samples: 177
• Classes: ['negative', 'positive']

Training Procedure

Preprocessing

• Image resizing to (224, 224)
• Normalization (pixel values scaled to [0,1])
• Data augmentation (rotation, shifting, flipping, zooming)

Training Hyperparameters

• Batch Size: 32
• Learning Rate: 0.001 (initial), 0.0001 (fine-tuning)
• Optimizer: Adam
• Loss Function: Binary Crossentropy
• Epochs: 50 (initial) + 20 (fine-tuning)

Evaluation Results

Metric	Value
Accuracy	0.8418
Precision	0.8509
Recall	0.8981
Loss	0.3919

How to Use

import tensorflow as tf
import numpy as np
from PIL import Image

# Load model
model = tf.keras.models.load_model('fine_tuned_phone_detection_model.h5')

def predict_phone_usage(image_path):
    # Preprocess image
    img = Image.open(image_path).convert("RGB")
    img = img.resize((224, 224))
    img_array = np.array(img) / 255.0
    img_array = np.expand_dims(img_array, axis=0)
    
    # Predict
    prediction = model.predict(img_array)[0][0]
    class_names = ['no_phone', 'using_phone']
    result = class_names[1] if prediction > 0.5 else class_names[0]
    confidence = prediction if prediction > 0.5 else 1 - prediction
    
    return result, confidence

Model Architecture

The model uses MobileNetV2 as base architecture with custom classification layers:

• Base Model: MobileNetV2 (pre-trained on ImageNet)
• Global Average Pooling
• Dropout (0.3)
• Dense Layer (128 units, ReLU activation)
• Batch Normalization
• Dropout (0.5)
• Output Layer (1 unit, Sigmoid activation)

License

MIT License