Abstract:Biosignal acquisition is key for healthcare applications and wearable devices, with machine learning offering promising methods for processing signals like surface electromyography (sEMG) and electroencephalography (EEG). Despite high within-session performance, intersession performance is hindered by electrode shift, a known issue across modalities. Existing solutions often require large and expensive datasets and/or lack robustness and interpretability. Thus, we propose the Spatial Adaptation Layer (SAL), which can be prepended to any biosignal array model and learns a parametrized affine transformation at the input between two recording sessions. We also introduce learnable baseline normalization (LBN) to reduce baseline fluctuations. Tested on two HD-sEMG gesture recognition datasets, SAL and LBN outperform standard fine-tuning on regular arrays, achieving competitive performance even with a logistic regressor, with orders of magnitude less, physically interpretable parameters. Our ablation study shows that forearm circumferential translations account for the majority of performance improvements, in line with sEMG physiological expectations.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the performance degradation between different sessions due to electrode position movement (electrode shift) in the application of biosignal sensor arrays. Although the performance within the same session is usually high, between different sessions, the performance drops significantly due to the influence of electrode shift. Existing solutions either require a large number of expensive data sets or lack robustness and interpretability. Therefore, the authors propose the **Spatial Adaptation Layer (SAL)**, which can be used as a pre - layer for any biosignal array model and learn a parameterized affine transformation at the input between two recording sessions. In addition, **Learnable Baseline Normalization (LBN)** is introduced to reduce baseline fluctuations. These methods aim to improve the performance between different sessions while maintaining the interpretability and efficiency of the model. ### Main Contributions 1. **Spatial Adaptation Layer (SAL)**: - **Function**: SAL can learn a parameterized affine transformation at the input between two recording sessions to adapt to electrode shift. - **Structure**: SAL is decomposed into separable transformations to improve interpretability. - **Advantage**: Compared with the traditional fine - tuning method, SAL uses fewer parameters and has higher interpretability. 2. **Learnable Baseline Normalization (LBN)**: - **Function**: LBN reduces the fluctuation of baseline activity by learning the bias of each biosignal channel during the adaptation process. - **Advantage**: LBN is more effective in reducing baseline fluctuations than traditional methods. ### Experimental Results - **Data Set**: The paper was tested on two publicly available high - density surface electromyography (HD - sEMG) gesture recognition data sets. - **Performance**: The experimental results show that SAL and LBN outperform the standard fine - tuning method in performance between different sessions, and can achieve performance comparable to complex models even when using a logistic regressor. - **Ablation Study**: Further ablation studies show that the circumferential translation of the forearm contributes the most to performance improvement, which is consistent with the physiological expectation of sEMG. ### Conclusion SAL and LBN can significantly improve the performance of biosignal sensor arrays between different sessions while maintaining the interpretability and efficiency of the model. These methods are particularly suitable for application scenarios that require rapid adaptation to new sessions. Future research will further explore the application potential of these methods in more practical one - dimensional sensor arrays.

Spatial Adaptation Layer: Interpretable Domain Adaptation For Biosignal Sensor Array Applications

Unsupervised Domain Adaptation by Causal Learning for Biometric Signal-based HCI

Surface EMG-Based Inter-Session Gesture Recognition Enhanced by Deep Domain Adaptation.

Semg-Based Gesture Recognition with Embedded Virtual Hand Poses and Adversarial Learning.

Physics-informed and Unsupervised Riemannian Domain Adaptation for Machine Learning on Heterogeneous EEG Datasets

Unsupervised Domain Adaptation for Activity Recognition Across Heterogeneous Datasets

Adversarial Domain Adaptation for Stable Brain-Machine Interfaces

Unsupervised Domain Adaptation for Inter-Session Re-Calibration of Ultrasound-Based HMIs

Learning-Based Regularization for Cardiac Strain Analysis via Domain Adaptation

Domain Adaptation with Self-Guided Adaptive Sampling Strategy: Feature Alignment for Cross-User Myoelectric Pattern Recognition

A Novel Hybrid Unsupervised Domain Adaptation Method for Cross-subject Joint Angle Estimation from Surface Electromyography

Tackling Electrode Shift In Gesture Recognition with HD-EMG Electrode Subsets

Reliable Source-Free Domain Adaptation for Cross-User Myoelectric Pattern Recognition

Align and Pool for EEG Headset Domain Adaptation (ALPHA) to Facilitate Dry Electrode Based SSVEP-BCI.

Electromyography Based Cross-Subject Limb Angle Estimation via Hierarchical Spiking Attentional Feature Decomposition Network

Self-supervised Representation Learning Framework for Remote Physiological Measurement Using Spatiotemporal Augmentation Loss

A Robust and Real-Time Framework of Cross-Subject Myoelectric Control Model Calibration via Multi-Source Domain Adaptation

Inter-Subject Domain Adaptation for CNN-Based Wrist Kinematics Estimation Using sEMG

A Novel Active Learning Framework for Cross-Subject Human Activity Recognition from Surface Electromyography

Cross-Subject Lifelong Learning for Continuous Estimation From Surface Electromyographic Signal

Learning Shared Semantic Information from Multimodal Bio-signals for Brain-Muscle Modulation Analysis