BrailleNet: Custom Light Weight Convolutional Neural Network (CNN) for Braille Character Classification

Introduction

BrailleNet is a convolutional neural network (CNN) designed for the classification of Braille characters. This repository includes the code for training the model using the Braille Dataset and provides an overview of the network architecture.

Dataset

1. The Braille Dataset is organized in the following structure:

   dataset/
       └── Braille Dataset/
           └── Braille Dataset/
               ├── a1.JPG
               ├── a2.JPG
               ├── ...
               └── z9.JPG
   

For more details or to explore the dataset, visit the Braille Character Dataset on Kaggle.

Model Architecture

The CNN architecture includes separable convolutional layers, batch normalization, and dropout for robust feature extraction. The model is trained on Braille characters with data augmentation.

Code

Here is a Python file called 'brailleNet_CNN.py' is used to train the model for Braille Character Classification.

Evaluation

After training, the model is evaluated on a validation set. Below is a summary of the model evaluation:

Metric	Value
Accuracy	97.44%
Recall	96.78%
F1-Score	96.81%
Precision	95.67%

Requirements

• Python 3.x
• TensorFlow
• Keras
• Matplotlib (for visualization)

Feel free to customize the model architecture or experiment with hyperparameters based on specific requirements.

License

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

Contact

For inquiries or feedback, feel free to contact us at msayeedi212049@bscse.uiu.ac.bd