Enhancing BISINDO Recognition Accuracy Through Comparative Analysis of Three CNN Architecture Models

This paper aims to develop an effective and efficient system for hand gesture recognition of Indonesian Sign Language (BISINDO) using a comparative analysis of three CNN architectures. The Python programming language is used as the development environment, together with the free edition of Google Colab. Google Colab provides a free online platform for coding and experimentation. PyTorch is used as a framework for training models, exploiting its Python programming language capabilities. and comprised 8,792 pictures of 200 by 200 pixel dimensions, divided A CNN-based technique is used, consisting of convolutional layers with ReLU activation, a comparison of three possible CNN designs from the detected image, and max pooling to constantly minimize dimensions, parameters, and computations within the network. The acquired findings show the CNN model is accurate after training and testing, with different accuracy rates.