Deep Learning Mastery through Network Ensemble Fusion for Indian Sign Language Recognition

Main Article Content

Priya K

Abstract

Indian Sign Language (ISL) serves as the primary means of communication for individuals with disabilities in India, yet it remains largely unfamiliar to the general population, hindering effective communication between these individuals and the wider community. Deep learning methodologies, crucial for tasks such as image classification and natural language processing, prioritize accuracy. Ensemble networks, which combine predictions from multiple models, present a promising avenue for enhancing accuracy in such applications. This research rigorously investigates various strategies for implementing ensemble networks and evaluates their efficacy in the domain of ISL recognition. The dataset utilized in this study comprises a comprehensive collection of ISL gesture images, encompassing the alphabet (A-Z) and numeric digits (0-9). Leveraging the robust capabilities of the MediaPipe framework, these images are organized in CSV format for efficient processing. Through extensive experimentation, we systematically examine the performance of ensemble approaches in comparison to individual models. Our findings consistently demonstrate that ensemble techniques exhibit superior performance over standalone models, showcasing a consistent trend towards improved accuracy. We have achieved around 96% for the average ensemble, whereas individually model was giving 95%, 89%, 93% respectively. We also attempted for weighted average ensemble, and we achieved around 97% which outperformed the individual models. This study contributes valuable insights into the utilization of ensemble networks for ISL recognition, highlighting their potential to bridge communication gaps between individuals with disabilities and the broader community.

Article Details

Section
Articles