Bangla Sign Language Classification with Deep Learning

BdSL (Bangla Sign Language) is a language in its own right, with its own grammar, syntax, and rules for word order. There are 36 alphabets in the Bangla sign language (BdSL), some of which have pretty similar characters. In our study, we use a convolutional neural network (CNN) method combined with Image Processing to classify Bangla sign alphabets and digits. We evaluated the performance of our scratched CNN model to pre-trained VGG-16 and model using the transfer learning method using our self-created dataset and IsharaLipi digits dataset. Our proposed model predicts digits with 99.38% accuracy and letters with 97% accuracy. In comparison to existing pre-trained models, our model needs extremely less computing resources and produces considerably superior accuracy outcomes.