Optimization of the CNN Model for Hand Sign Language Recognition Using Adam Optimization Technique

To communicate with someone, hand signs prove to be more effective than just words to emphasize and structure the conversation and give more clarity. Some machines, like a driving car or a robot, are also operated by hand gestures as they are more convenient to interpret, but devising the recognition of hand gesture system is an extremely arcane commission since the model should be clever enough to recognize the hand gestures in distinct positions and orientation. In this research, the convolution neural network (CNN) model of deep learning (DL) is utilized for training the hand sign language image dataset. In addition to this, the Adam optimization technique, which is known to leverage the adaptive learning technique for figuring out the learning rate for every parameter, is utilized here for determining the optimized values of hyperparameters. The dataset is taken from GitHub and is contributed by the Turkey Ankara Ayrancı Anadolu High School students. The dataset consists of 218 hand sign sample images of each of the ten digits ranging from 0–9. After determining the optimized values of various hyperparameters, the proposed framework is then validated upon the validation dataset. The training and validation loss over the optimized number of epochs comes out to be 0.021 and 0.064, respectively. Conclusively, the investigational outcomes depict that the anticipated optimized CNN model displays an increased accuracy of 98%.