| name | efficient-robust-snn-semg-fatigue-detection |
| description | A skill for implementing efficient and robust spiking neural networks for surface electromyography (sEMG) based muscle fatigue detection, based on arXiv:2607.11065 |
| license | Complete terms in LICENSE.txt |
Efficient and Robust Spiking Neural Networks for sEMG-Based Muscle Fatigue Detection
Overview
This skill provides a workflow for implementing efficient and robust spiking neural networks (SNNs) for muscle fatigue detection using surface electromyography (sEMG) signals, based on the research presented in arXiv:2607.11065.
Authors: Kaiwen Tang, Jiaqi Dong, Zhanglu Yan, Weng-Fai Wong
Published: 2026-07-13T04:05:13Z
Abstract: Detecting muscle fatigue via surface electromyography (sEMG) is essential for applications in sports, rehabilitation, and wearable health monitoring. Accurate and timely detection of fatigue is crucial for preventing injuries, optimizing physical performance, and ensuring user safety during prolonged activity. However, existing deep learning models are often unsuitable for this task due to their high computational cost and dependence on large-scale data. In this work, we propose an energy-efficient framework for muscle fatigue detection based on Spiking Neural Networks (SNNs), which exploit sparse, event-driven computation and temporal modeling. We further introduce a quantization-compatible training scheme (SDH) that combines multiple regularization terms to improve robustness under noisy conditions. Evaluated on two public sEMG datasets against a broad set of baselines and under seven noise conditions including physically motivated perturbations, our quantized SNNs match or exceed strong baselines while remaining more stable under diverse noise and reducing estimated energy consumption by up to 201.77x. These results demonstrate the framework's strong potential for real-time deployment in low-power wearable systems.
Methodology
The approach involves:
- Preprocessing sEMG signals (filtering, normalization, segmentation)
- Encoding time-series data into spike trains using temporal encoding schemes
- Training a spiking neural network with a quantization-compatible training scheme (SDH)
- Evaluating robustness under various noise conditions
- Optimizing for low-power deployment in wearable systems
Key Components
Spiking Neural Network Architecture
- Input layer: Encoded sEMG features
- Hidden layers: Leaky Integrate-and-Fire (LIF) neurons with surrogate gradients
- Output layer: Classification layer for fatigue levels (e.g., non-fatigued, fatigued)
SDH Training Scheme
The Stochastic Differential Homeostasis (SDH) method combines:
- Membrane potential regularization
- Spike rate regularization
- Synaptic weight normalization
- To improve robustness against noise and quantization effects
Workflow
Step 1: Data Preparation
- Collect sEMG data from target muscles during fatiguing contractions
- Apply bandpass filtering (typically 20-450 Hz) to remove noise
- Normalize signals to zero mean and unit variance
- Segment into fixed-width windows (e.g., 200ms with 50% overlap)
- Extract features (e.g., MAV, RMS, WL) or use raw signals for temporal encoding
Step 2: Spike Encoding
Convert preprocessed sEMG features into spike trains using:
- Temporal Encoding: Map feature values to spike latencies (lower value = earlier spike)
- Rate Encoding: Map feature values to firing rates over a fixed time window
- Phase Encoding: Use oscillatory phases to encode information
Step 3: Network Training
- Initialize SNN with random weights
- For each training epoch:
- Present encoded sEMG samples to the network
- Compute post-synaptic potentials and generate output spikes
- Calculate loss (e.g., cross-entropy) between predicted and true fatigue labels
- Update weights using surrogate gradient descent
- Apply SDH regularization terms
- Validate on held-out set to prevent overfitting
Step 4: Robustness Evaluation
Test the trained SNN under various noise conditions:
- Additive Gaussian noise (different SNR levels)
- Baseline wander
- Motion artifacts
- Electrode displacement simulations
Step 5: Deployment Optimization
- Quantize weights and activations to reduce memory footprint
- Optimize for neuromorphic hardware or low-power microcontrollers
- Implement event-driven processing to minimize energy consumption
Pitfalls and Tips
Common Pitfalls
- Insufficient temporal resolution: Encoding too coarsely loses dynamic sEMG characteristics
- Overfitting to noise: SDH regularization is crucial for robustness
- Hardware mismatch: SNN parameters must match target neuromorphic substrate
Tips for Success
- Validate encoding schemes on raw sEMG to ensure spike trains preserve discriminative information
- Start with small networks (2-3 hidden layers) and scale up as needed
- Use surrogate gradients like the fast sigmoid or straight-through estimator
- Monitor spike rates to avoid saturation or silence
References
- Primary paper: arXiv:2607.11065 - Efficient and Robust Spiking Neural Networks for sEMG-Based Muscle Fatigue Detection
- Related work: See references/ directory for detailed paper summary
Activation Keywords
- efficient-robust-snn-semg-fatigue-detection
- snn
- semg
- muscle fatigue
- spiking neural network