SCAI-net: spatial-channel attention Inception-net for Hand Gesture Recognition for deaf community medical issues

Hand gestures play a vital role in facilitating effective sign language communication, bridging the gap between the deaf and hearing communities. However, many existing hand gesture recognition (HGR) techniques struggle to accurately classify gestures with similar patterns. To address this challenge, SCAI-net is presented as an innovative HGR framework that integrates spatial and channel attention mechanisms into a modified Inception V3 architecture. In this approach, gestures are first processed using a pre-trained Inception V3 model, optimized to reduce computational overhead while enhancing feature extraction. Attention mechanisms are then applied to the model’s output to focus on subtle, discriminative features, significantly improving recognition accuracy for ambiguous gestures. Evaluation on the publicly available double-hand Indian Sign Language (ISL) dataset achieved a Top-1 accuracy of 99.46%, marking a 5.23% improvement over current state-of-the-art methods. This HGR system not only improves gesture recognition accuracy but also facilitates more effective communication between deaf individuals and healthcare providers through a user-friendly graphical user interface (GUI) prototype, enabling clearer understanding of medical conditions.