Face Mask Detection using Convolutional Neural Network (CNN)

Overview

This project implements a Face Mask Detection model using a custom Convolutional Neural Network (CNN). The model classifies images into three categories:

1. People wearing masks
2. People not wearing masks
3. People wearing masks but in an incorrect manner

The dataset used for training and testing is obtained from Kaggle, specifically the ["Face Mask Detector (mask, not mask, incorrect mask)" dataset].

Usage

1. Installation

To run the notebook locally, follow these steps:

# Clone the repository
git clone https://github.com/your-username/face-mask-detection.git
cd face-mask-detection

# Install the required dependencies
pip install -r requirements.txt

2. Dataset

Download the dataset from Kaggle using the Kaggle API key. Upload your Kaggle API key (kaggle.json) as mentioned in the notebook.

# Download the dataset using Kaggle API
!kaggle datasets download -d spandanpatnaik09/face-mask-detectormask-not-mask-incorrect-mask

# Unzip the downloaded dataset
!unzip face-mask-detectormask-not-mask-incorrect-mask.zip -d dataset

3. Training Images

Below are sample images from the training dataset:

4. Running the Notebook

Open the Jupyter notebook Face_Mask_Detection.ipynb in your local environment and run the code. Make sure to adjust any paths or configurations if needed.

Data Augmentation Parameters

The following data augmentation techniques are applied to the training dataset:

• Rescale: Pixel values are rescaled to the range of [0, 1].
• Rotation: Random rotation of images by up to 20 degrees.
• Width Shift: Random horizontal shift of images by up to 20% of the total width.
• Height Shift: Random vertical shift of images by up to 20% of the total height.
• Zoom: Random zooming into images by up to 20%.
• Horizontal Flip: Randomly flip images horizontally.

Model Architecture

The Face Mask Detection model is built using TensorFlow and Keras. The CNN model architecture consists of several layers, each serving a specific purpose in the image classification task.

• Convolutional Layers: Three convolutional layers with 64 filters and a kernel size(2x2).
• MaxPooling Layers: Three max-pooling layers with a pool size of 2x2.
• Flatten Layer: Flattens the output from the convolutional layers into a one-dimensional array.
• Dense Layers: Two dense layers with ReLU activation functions. The first dense layer consist of 128 neurons, providing a high-level representation of the learned features. The second dense layer, with 3 neurons and softmax activation, produces the final probability distribution for the three classes ('with_mask', 'without_mask', 'incorrect_mask').
• Dropout Layer: Introduces dropout regularization with a rate of 0.5 to prevent overfitting during training.

5. Results

Below are some predicted label images from the test dataset:

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

• Dataset: Face Mask Detector (mask, not mask, incorrect mask)
• TensorFlow and Keras for deep learning tools