Indian Sign Language Recognition – Deep Learning Models

📊 Overview

This project uses deep learning models to classify Indian Sign Language (ISL) gestures into their corresponding labels. The objective is to recognize ISL gestures using two powerful models: ResNet50 and MobileNetV2. The project leverages transfer learning, using pre-trained models to reduce training time and improve classification accuracy.

📁 Repository Contents

• Indian_Sign_Langauge_Prediction.ipynb – Main script for training and evaluation
• README.md – Project documentation

🧩 Objectives

• Build a deep learning model for recognizing ISL gestures.
• Use pre-trained models like ResNet50 and MobileNetV2 for transfer learning.
• Evaluate the models based on accuracy, both on training and testing datasets.

📌 Key Features

• Transfer Learning: Utilizes pre-trained models (ResNet50 and MobileNetV2) to enhance classification performance with fewer training data.
• Data Augmentation: Augments the training data with rotations, shifts, zooms, and flips to make the model more robust.
• Model Evaluation: Evaluates the models on both training and test datasets, reporting accuracy.
• Visualization: Visualizes the training and validation accuracy curves for both models.

📈 Key Insights

• ResNet50 achieved 88.65% accuracy on the test set.
• MobileNetV2 achieved 87.23% accuracy on the test set.
• Both models performed well, with training accuracies above 98%.

🛠️ Tools Used

• Python 3.x
• TensorFlow 2.x
• Keras
• OpenCV
• Matplotlib
• scikit-learn
• KaggleHub

📂 Datasets Used

• Indian Sign Language Dataset: A collection of images of ISL gestures, categorized by letters, numbers, and other common gestures.

Note: The dataset is preprocessed for resizing images, encoding labels, and applying augmentation techniques.

🔄 How to Use

1. Clone or download this repository.
2. Install the required dependencies:

   pip install tensorflow opencv-python matplotlib scikit-learn kagglehub

3. Run the Indian_Sign_Langauge_Prediction.ipynb script to train the models.
4. The models will be trained on the dataset and display evaluation results, including training and test accuracy.
5. Use the plots generated by matplotlib to visualize model performance.

📌 Model Details

ResNet50

• ResNet50 is a deep residual network that uses skip connections to solve the vanishing gradient problem in deep networks. This architecture is well-suited for image classification tasks as it can capture hierarchical features effectively.
• Pre-trained on ImageNet, the model is fine-tuned on the ISL dataset to classify gestures.
• Fine-tuning was performed starting from layer 80 (pre-trained layers frozen).

MobileNetV2

• MobileNetV2 is an efficient convolutional neural network architecture designed for mobile and edge devices. It uses depthwise separable convolutions to reduce computation while maintaining accuracy.
• The model is pre-trained on ImageNet and fine-tuned on the ISL dataset for gesture classification.
• Fine-tuning was done from layer 50 to allow the model to adapt to the new dataset.

📌 Use Cases

• Sign Language Interpreters: Automate interpretation of sign language in real-time applications.
• Educational Institutions: Develop learning tools for students to learn sign language.
• Tech Companies: Integrate gesture recognition in apps and systems for accessibility.

🚀 Future Enhancements

• Experiment with other models like VGG16 or InceptionV3 to compare performance.
• Implement a real-time sign language recognition system using a webcam.
• Integrate the model into a mobile or web app for broader accessibility.

📬 Contact

For queries or collaboration:

GitHub: ritup04
Email: ritupal1626@gmail.com