SSR Using Portable sEMG Acquisition System with Electrode Layout Optimization

Surface Electromyography (sEMG) based Silent Speech Recognition (SSR) holds significant potential for medical and specialized scenarios. However, existing studies predominantly rely on bulky wired single electrode sEMG acquisition systems with arbitrarily placed electrodes, resulting in limitations related to portability, complex deployment procedures, and signal redundancy. In this study, we addressed these challenges by developing a lightweight wireless sEMG acquisition system equipped with a flexible array of e-skins. We propose a three-layer laminated electronic skin based on flexible PCB. Its shape is designed with electrodes separated from each other to improve the fitting performance when the skin deforms. Finally we optimizing the electrode placement using Principal Component Analysis (PCA) and Genetic Algorithm (GA) with nonlinear channel compensation. Our experimentation demonstrated a high classification accuracy of 93.62% when using all 32 channels. Moreover, the 16-channel classification accuracy achieved using PCA and GA was 91.49% and 97.82%, respectively. An optimal selection method was identified, where 11 channels selected using GA yielded a classification accuracy of 95.74%, striking a balance between accuracy and channel usage. Furthermore, our analysis revealed the heightened significance of electrodes in muscle groups near the corners of the mouth, and proposed the feasibility of acquiring unilateral sEMG signals for symmetricalmuscle groups. Our acquisition system holds promise for practical implementation in silent speech recognition scenarios, offering potential for daily use. The three-layer e-skin electrode design, along with the channel optimization method and electrode position selection strategy can serve as a valuable reference for the development of similar e-skin electrodes.