Advancing Communication: Deep Learning Innovations in Bengali Sign Language Recognition

Sign language is vital for the Deaf to communicate with others in society to express themselves in terms of ideas, feelings, and thoughts which involve both the bodily gestures and the aspects of the visually perceived environment. The emergence of sign language detection system is crucial in order to improve the quality of life and communication of the deaf people. This research seeks to fill this gap by investigating on Bengali Sign Language which is a specific linguistic and cultural mode of communication. This research aims to contribute to the knowledge area of Bengali Sign Language Recognition using state-of-the-art Convolutional Neural Networks (CNN) such as InceptionResNetV2 and MobileNetV2. These include ResNet50V2, DenseNet121, InceptionV3, and Xception. Using three different datasets, which include only digit, word, and letter images, the models have gone through the processes of training and testing phases, and we used 1630 digit images, 1200 word images, and 4430 letter images to get better accuracy of the models. From the performance evaluation, the potential of some models was emphasized, including DenseNet121 and MobileNetV2, which have shown high accuracy repeatedly and the ability to adapt to different datasets at all the mentioned platforms. For digits, words, and letters, we got 99.07%, 96.67%, and 97.29% accuracy. These outcomes combined with the ROC curves are indicative of the effectiveness of the chosen models in studying the minute details within signs and gestures, and thus they promote the progress of this area and the goal of making Bengali Sign Language Interpreting more accessible to the public. The present study can therefore serve as groundwork for subsequent research in the field of the study. Expand in the field of sign language recognition and thus promote better communication. interactive forms for those people using Bengali Sign Language from different fields of life experience.