On Optimizing Electrode Configuration for Wrist-Worn sEMG-Based Thumb Gesture Recognition

Abstract

Thumb gestures provide an effective and unobtrusive input modality for wearable and always-available human-machine interaction. Wrist-worn surface electromyography (sEMG) has emerged as a promising approach for compact and wearable human-machine interfaces. However, compared to forearm sEMG, the impact of electrode configuration on wrist-based decoding performance remains understudied. We systematically investigated electrode configuration strategies for wrist-based thumb-movement recognition using high-density (HD) and low-density (LD) sEMG measurement systems. We considered factors such as muscle region, reference scheme, channel count, and spatial density of the electrode. Experimental results show that 1) extensor-side electrodes outperform flexor-side electrodes (HD: 0.871 vs. 0.821; LD: 0.769 vs. 0.705); 2) monopolar recordings consistently outperform bipolar configurations (15 channel with HD monopolar vs. LD bipolar: 0.885 vs. 0.823); and 3) increasing channel count enhances performance, but exhibits diminishing returns. We further show that electrode spatial distribution introduces a trade-off between spatial coverage and compactness. The findings suggest that the effectiveness of wrist-worn sEMG systems depends less on the deployment of a large number of electrodes in a broad sensing area and more on the optimization of electrode placement and the referencing scheme. This work provides practical guidelines for developing efficient wrist-worn sEMG-based gesture recognition systems.

Figures (5)

• Figure 1: Overview of proposed study illustrating the experimental setup and the deep learning framework for thumb gesture recognition. (a) Trigno Maize sensor, a high-density sEMG electrode grid including 16 channels. (b) Placement of two sensor grids on the extensor and flexor sides of the right wrist, with the central column aligned with the forearm midline. (c) Illustration of the thumb movements. (d) Example of recorded sEMG signals. (e) Experimental setup showing synchronized sEMG acquisition and multi-camera recording. (f) Camera views used for hand motion capture. (g) Reconstructed three-dimensional hand kinematics obtained from multi-view camera recordings. (h) Visualization of hand joint landmarks. (i) CNN architecture.
• Figure 2: Electrode spatial maps derived from the Maize HD grid showing three levels of spatial sampling density. An eight-electrode configuration is illustrated. High-density maps (top) have shared-edge connectivity (red solid liens). Medium-density maps (middle) have shared-corner connectivity (blue dashed lines) only, and Low-density maps (bottom) consist of spatially isolated electrodes.
• Figure 3: Thumb-movement classification performance under different electrode configurations. (a) Performance comparison between extensor (Ext.), flexor (Fle.), and combined electrode placements for Maize sensor. (b) Corresponding placement comparison for Quattro sensor. (c) Comparison between monopolar and bipolar configurations using Maize and Quattro sensors across subjects. (d) Effect of random channel reduction on classification accuracy for Maize and Quattro sensors.
• Figure 4: Electrode importance maps for thumb-movement classification using the Maize sensor. (a)--(f) show normalized integrated gradients (IG) attribution distributions obtained from the CNN model for different gestures. Attribution scores were computed with respect to raw sEMG inputs, normalized within each subject, and averaged across subjects. The color scale (blue to red) indicates increasing contribution of each electrode to the CNN model output.
• Figure 5: Effect of electrode spatial density under four-, six-, and eight-channel configurations using the Maize sensor. (a)--(c) relationship between classification accuracy and spatial distance ($\mathrm{Dist}$). (d)--(f) The relationship between the figure of merit (FOM) and electrode density. Each point represents one randomly sampled electrode configuration.