DenseNet121 with Harris Hawks Optimization: A Novel Deep Learning Approach for Arabic Sign Language Recognition

This paper presents a novel approach for Arabic Sign Language (ArSL) recognition by leveraging advanced deep learning models with optimization techniques. Several architectures, including CNNs, EfficientNet, ResNet50, and DenseNet, are evaluated and integrated with Harris Hawks Optimization (HHO) to enhance performance. Our DenseNet121 model, optimized with HHO, achieved state-of-the-art accuracy, surpassing previous models. Using the ArASL2018 dataset of 54,049 images, models were evaluated on metrics such as accuracy, precision, recall, and F1-score. The DenseNet121 and HHO model achieved an impressive 99.79 percent, demonstrating its potential for assistive technologies and human-computer interaction.