DeepBdSL: A Comprehensive Assessment of Deep Learning Architectures for Multiclass Bengali Sign Language Gesture Recognition

Bengali Sign Language (BdSL) is a critical communication medium for the Deaf community, yet significant challenges persist in bridging communication barriers with the hearing population. Early and accurate recognition of BdSL gestures is essential for fostering inclusion and understanding between the Deaf and hearing communities. This study comprehensively evaluates deep learning models to recognize BdSL gestures, focusing on the effectiveness of architectures like VGG16, ResNet50, and a custom Convolutional Neural Network (CNN). Using a dataset of 11,172 images representing 38 distinct BdSL gestures sourced from Kaggle, the proposed CNN model was trained to optimize recognition accuracy. The model’s performance was evaluated using precision, recall, and F1-score, achieving an accuracy of 92.33%. This surpasses traditional machine learning approaches and other pre-trained models, demonstrating the model’s superiority in recognizing complex BdSL gestures. The results highlight the model’s potential for developing accessible communication tools, promoting better interactions between Deaf and hearing communities, and advancing automated BdSL recognition systems. These findings provide a foundation for further research in this area.