Model Penerjemah Bahasa Isyarat SIBI Statis Berbasis Convolutional Neural Network

Abstract

This study addresses the communication gap between deaf and hearing communities by developing an optimal sign language recognition system for Indonesian Sign Language System (SIBI) static gestures. A comprehensive comparative analysis was conducted on four VGG architecture variants (VGG-11, VGG-13, VGG-16, and VGG-19) using a dataset across 10 SIBI word classes. The research employed systematic methodology including data extraction from video sources, preprocessing with augmentation techniques, model training over 25 epochs, and comprehensive evaluation using accuracy, precision, recall, and F1-score metrics. Results demonstrate that VGG-16 achieves superior performance with 83.4% accuracy, 85.2% precision, 83.4% recall, and 82.9% F1-score, establishing optimal balance between model complexity and generalization capability. The study reveals diminishing returns phenomenon in VGG-19 despite increased architectural complexity. Computational efficiency analysis shows VGG-11 provides highest efficiency score (10.46 GFLOPs) while VGG-16 maintains optimal accuracy-efficiency trade-off. These findings provide crucial insights for developing effective assistive technology solutions that bridge communication barriers for the Indonesian deaf community.