The limited availability of information regarding Indonesian Sign Language (BISINDO) alphabet recognition poses a significant challenge to effective communication between individuals with normal hearing and deaf individuals. This research aims to address this issue by developing a Convolutional Neural Network (CNN) model using the DenseNet-201 architecture to recognize the BISINDO alphabet and evaluate its accuracy. Additionally, this study seeks to determine the accuracy level, design a BISINDO alphabet classification system employing CNN and DenseNet-201, and develop an initial stage that serves as a core component for the broader recognition of the entire BISINDO dictionary. The model is designed to classify classes based on hand shapes and patterns through several stages, including data collection and splitting, image resizing, data augmentation, and the application of the DenseNet-201 architecture, which involves convolutional, max-pooling, ReLU, dropout, and dense layers. The model training process was conducted over 22 epochs with a batch size of 32, and the system implementation utilized the Django framework. The results show that the best model was achieved using the DenseNet-201 architecture, a batch size of 32, and 22 epochs of training, yielding an accuracy of 96%. This study makes a significant contribution to enhancing the effectiveness of communication between hearing and deaf individuals, while offering an innovative technological solution for sign language recognition.
ResearcherCollab
