SASG-DA: Sparse-Aware Semantic-Guided Diffusion Augmentation For Myoelectric Gesture Recognition

TL;DR

This paper tackles the data scarcity and overfitting challenges in sEMG-based gesture recognition by introducing SASG-DA, a diffusion-driven data augmentation framework. SASG-DA combines Semantic Representation Guidance (SRG) for faithful conditioning, Gaussian Modeling Semantic Sampling (GMSS) for flexible diversity, and Sparse-Aware Semantic Sampling (SASS) to explicitly expand coverage into underrepresented semantic regions. Through extensive experiments on Ninapro DB2/DB4/DB7 across three backbones, SASG-DA achieves consistent improvements over state-of-the-art augmentation methods, demonstrating gains in accuracy, fidelity (FID/CAS), and dataset coverage while mitigating overfitting. The approach shows practical potential for improving HMI and prosthetic control systems and offers a scalable blueprint for semantically guided, diversity-aware diffusion augmentation in time-series biosignal tasks.

Abstract

Surface electromyography (sEMG)-based gesture recognition plays a critical role in human-machine interaction (HMI), particularly for rehabilitation and prosthetic control. However, sEMG-based systems often suffer from the scarcity of informative training data, leading to overfitting and poor generalization in deep learning models. Data augmentation offers a promising approach to increasing the size and diversity of training data, where faithfulness and diversity are two critical factors to effectiveness. However, promoting untargeted diversity can result in redundant samples with limited utility. To address these challenges, we propose a novel diffusion-based data augmentation approach, Sparse-Aware Semantic-Guided Diffusion Augmentation (SASG-DA). To enhance generation faithfulness, we introduce the Semantic Representation Guidance (SRG) mechanism by leveraging fine-grained, task-aware semantic representations as generation conditions. To enable flexible and diverse sample generation, we propose a Gaussian Modeling Semantic Sampling (GMSS) strategy, which models the semantic representation distribution and allows stochastic sampling to produce both faithful and diverse samples. To enhance targeted diversity, we further introduce a Sparse-Aware Semantic Sampling (SASS) strategy to explicitly explore underrepresented regions, improving distribution coverage and sample utility. Extensive experiments on benchmark sEMG datasets, Ninapro DB2, DB4, and DB7, demonstrate that SASG-DA significantly outperforms existing augmentation methods. Overall, our proposed data augmentation approach effectively mitigates overfitting and improves recognition performance and generalization by offering both faithful and diverse samples.

Figures (10)

• Figure 1: Overfitting remains a persistent challenge in sEMG-based gesture recognition due to the scarcity of informative data. We alleviate this by generating faithful and diverse samples through diffusion-based augmentation. Our sparse-aware sampling strategy further encourages generation in underrepresented regions, effectively expanding the data distribution.
• Figure 2: Overview of the proposed SASG-DA approach based on diffusion model. (b) SASG-DA enhances generation faithfulness by introducing semantic representations to guide diffusion training. (c) During inference, by (d) modeling semantic representation distribution, SASG-DA performs (e) Sparse-Aware Semantic Sampling to target the sparse regions in the semantic space and expand the data distribution. (f) The optimized sparse semantic representation are then used as condition inputs for diffusion reverse and generate synthetic sEMG signals for downstream model training.
• Figure 3: Sparse-Aware Semantic Sampling. Candidate sparse features (blue) are optimized in the modeled semantic space using two potential functions. (a) The sparsity potential applies a repulsive force between candidate and reference features (orange), pushing candidates toward sparse regions. (b) The diversity potential encourages mutual repulsion among candidates to spread out within the semantic space.
• Figure 4: Classification accuracy (%) ablation results for all the subjects from (a) DB7 and (b) DB4.
• Figure 5: Distribution visualizations of the original (blue) and generated (red) samples for subject 1 from (a) DB7, (b) DB4 and (c) DB2. Different shades of each color denote different gesture classes.