Sign language recognition based on CNN with optimized activation function

Sign language is a non-verbal language that is used mostly by individuals with hearing impairments, yet it is not a common language among people without hearing disabilities. A Convolutional Neural Network (CNN) model to recognize sign language is developed to overcome the barrier of communication between individuals. In this research, the model is trained with altered first, second, and third dense layers; their activation functions are changed from ReLU to other activation functions to find out the best activation function to create the most accurate model. The new activation functions chosen in our research include Sigmoid, Tanh, Softmax, Softplus, ThresholdedReLU, ELU, and PReLU. The comparison of the accuracy of models trained with different activation functions is provided.