An Efficient Deep Learning Model for Static ASL Gesture Recognition in Resource-Constrained Environments

This research presents an efficient and lightweight CNN model for recognizing static American Sign Language (ASL) utilizing RGB images that cover 36 gesture categories. The model is built from the beginning using L2 regularization and dropout to avoid overfitting. It is improved using a method called stochastic gradient descent (SGD), along with a system to adjust the learning rate. Using 200 × 200 color images, it achieves a test accuracy of 95.83%, which is better than many existing methods, with approximately 1.8 billion floating-point operations (FLOPs). Its efficiency and accuracy make it suitable for devices with constrained resources.