SparseEMG: Computational Design of Sparse EMG Layouts for Sensing Gestures

TL;DR

SparseEMG is a data-driven tool that optimizes electrode placement for EMG gesture recognition, significantly reducing electrode count while maintaining high accuracy across diverse datasets and hardware setups.

Contribution

This work introduces the first systematic analysis of electrode selection and classifier impact on EMG gesture recognition, and presents SparseEMG, a tool for designing sparse electrode layouts.

Findings

Permutation Importance with Random Forest reduces electrodes by 53.5%.

SparseEMG layouts are transferable across users with minimal performance loss.

The tool supports over 50 gestures and is validated in real-world applications.

Abstract

Gesture recognition with electromyography (EMG) is a complex problem influenced by gesture sets, electrode count and placement, and machine learning parameters (e.g., features, classifiers). Most existing toolkits focus on streamlining model development but overlook the impact of electrode selection on classification accuracy. In this work, we present the first data-driven analysis of how electrode selection and classifier choice affect both accuracy and sparsity. Through a systematic evaluation of 28 combinations (4 selection schemes, 7 classifiers), across six datasets, we identify an approach that minimizes electrode count without compromising accuracy. The results show that Permutation Importance (selection scheme) with Random Forest (classifier) reduces the number of electrodes by 53.5\%. Based on these findings, we introduce SparseEMG, a design tool that generates sparse electrode layouts based on user-selected gesture sets, electrode constraints, and ML parameters while also predicting classification performance. SparseEMG supports 50+ unique gestures and is validated in three real-world applications using different hardware setups. Results from our multi-dataset evaluation show that the layouts generated from the SparseEMG design tool are transferable across users with only minimal variation in gesture recognition performance.