Deep Learning Based Recognition of Arabic Alphabet Sign Language ArASL: A Study with EfficientNetB3

Abstract

Sign language is a critical communication approach for the community of people with hearing and speech impairments . Humans need to be able to communicate. People who are unable to communicate verbally like the rest of humanity typically utilize sign language .The primary characteristics of signs in sign language are hand form, placement, movement, orientation, and non-manual elements. These people are facing significant obstacles in their lives, such as severe depression and unemployment as a result of these restrictions or impairments. Among the communication services they utilize is a sign language interpreter. However, the cost of hiring these interpreters makes a low-cost option necessary. As a result, society now urgently needs automatic sign language translation. The construction of image-based Arabic sign language (ArSL) identification systems has improved due to the accessibility and extensive usage of digital cameras on mobile phones.  and represent an opportunity for people with hearing disabilities to participate more in their communities. Their quality of life would be significantly impacted by the creation and deployment of a new system for the recognition of (ArSL). Consequently, a method was developed to translate the visual hand data set from Arabic sign language to written data. The objective of this work is to use the EfficientNetB3 deep learning model to achieve state-of-the-art performance in the classification of Arabic Alphabet Sign Language. The suggested method achieves an impressive test accuracy of 99.84% by utilizing transfer learning, data augmentation techniques, and a well selected dataset. The outcomes show how the model can be used in practical settings for things like real-time sign language interpreters and educational resources.