DEAH-Net: A Dual-Modal 3D-CNN Framework for sEMG and Acceleration-Based HMI

Recent advancements in human-machine interaction (HMI) have significantly enhanced applications in bionic robotics, intelligent prosthetics, augmented reality (AR), and virtual reality (VR). To address these challenges, we propose DEAH-Net, a dual-modal 3D convolutional neural network (3D-CNN) integrating sEMG and triaxial acceleration data. Data is acquired via a high-resolution ADS1298 analog-to-digital converter and MPU6050 sensors, managed by an NRF52832 microcontroller for preprocessing. DEAH-Net features a two-branch architecture combining Dense-Net with an Acceleration Information Enhancement Branch (AIEB), significantly improving gesture and force classification accuracy. Experiments demonstrate that DEAH-Net achieves 99.9951% accuracy in gesture recognition tasks, showing strong generality and robustness across various environments. DEAH-Net presents an efficient and practical solution for multimodal gesture recognition in HMI applications.