Scholarly output

Publications

IEEE-style references with research highlights and additional information.

PUBLISHED · IEEE XPLORE · 2026

Fatigue-Resistant s-EMG Hand Gesture Recognition with Lightweight Recurrent–Attention Networks

[1] A. Bagheri Tirtashi and M. Jahed, “Fatigue-Resistant s-EMG Hand Gesture Recognition with Lightweight Recurrent–Attention Networks,” in Proc. 34th International Conference on Electrical Engineering (ICEE), 2026, doi: 10.1109/ICEE70628.2026.11710601.

Research highlight: Studied whether compact sequence models with multi-head self-attention can maintain reliable sEMG-based gesture recognition when muscle-fatigue-like distortions alter signal amplitude and frequency content.

IEEE Xplore ↗DOI ↗

Read research highlights

Research question: How do compact low-channel sEMG gesture decoders respond to fatigue-like changes in signal amplitude and spectrum?

Methods: Eight healthy participants, six forearm sEMG channels, four hand/wrist gestures, and 250 ms signal windows. Compared classical feature-based models against CNN, GRU, and multi-head self-attention variants.

Key findings: MHA+CNN achieved 98.23% accuracy on clean recordings. MHA+GRU retained 65.19% accuracy at the strongest simulated amplitude attenuation and 84.58% at the strongest simulated spectral tilt.

Limitations: Muscle fatigue was simulated on non-fatigued signals; the results do not establish performance under naturally occurring fatigue or in clinical deployments.

ACCEPTED · AWAITING PUBLICATION · 2026

Robust EEG–sEMG Fusion for Hand Gesture Recognition Using Cross-Attention

[2] A. Bagheri Tirtashi and M. Jahed, “Robust EEG–sEMG Fusion for Hand Gesture Recognition Using Cross-Attention,” 33rd International Conference on Biomedical Engineering (ICBME), 2026, accepted for publication.

Research highlight: Proposed a cross-attention EEG–sEMG fusion model that explicitly learns relationships between neural and muscular representations, aiming to improve hand-gesture decoding when muscular signals are degraded.

Read research highlights

Research question: Can complementary EEG information improve sEMG hand-gesture recognition when muscular signals are degraded?

Methods: Synchronized EEG and six-channel sEMG from eight healthy participants performing four gestures. Used independent preprocessing, parallel CNN encoders, self-attention, and cross-attention for multimodal feature fusion.

Key findings: Cross-attention fusion achieved 92.45% clean-data accuracy, 78.05% under the strongest simulated amplitude attenuation, and 75.46% under the strongest simulated spectral tilt.

Limitations: The perturbations simulate selected fatigue-related signal changes rather than measuring physiological fatigue; clinical generalization remains untested.