Sensor Topology-Aware Three-Branch Fusion for sEMG Gesture Recognition

An Exploratory Study of Single Channel Surface Electromyography for Hand Gesture Classification

This study investigates the feasibility of classifying ten hand gestures using a single sEMG channel combined with lightweight machine learning architectures and demonstrates that combining time and frequency features with Pearson filtering and a compact NN can achieve up to 90 percent accuracy.

Multimodal Surface EMG Hand Gesture Recognition Using Query-Based Transformers for Prosthetic Control

EMG-CrossFormer is introduced, an end-to-end hybrid convolutional-transformer for seamless multimodal integration that improves sEMG-only decoding and that multimodal fusion substantially amplifies this benefit, underscoring the value of both design principles for complex hand gesture recognition.

Knowledge distillation for sEMG-based gesture recognition: enhancing wearable HMI systems with lightweight models

Objective. Gesture recognition using surface electromyography (sEMG) is a promising technology for wearable human–machine interaction systems. However, sEMG-based gesture classification models often suffer from performance degradation in cross-user scenarios due to the variability in individual physiological signals. To address this issue, large-scale deep learning algorithms with a great amount of parameters are employed to enhance model generalization and improve recognition performance across different users. However, the use of deep learning models on wearable devices is challenging due to their limited computational power and memory capacity. Methods. This study addresses this challenge through knowledge distillation, which compresses large teacher models into lightweight student models while maintaining performance. We trained deep teacher models, including DenseNet, InceptionV3, and VggNet, and distilled their parameters into compact convolutional neural network and long short-term memory (LSTM) models. The experiment was performed on the sEMG data of 30 subjects collected from a wrist-band electrode with 32 channels. Main Results. The results demonstrated that student models, particularly LSTM-based ones, achieved classification accuracy close to or even higher than their teacher models, with a highest accuracy improvement. Among the teacher–student combinations, the DenseNet121-LSTM architecture achieved the highest classification accuracy. The relationship between computational complexity floating-point operations (FLOPs) and model performance was also analyzed, showing that the distilled models can effectively approximate high-FLOP models. Significance. This research enhances the feasibility of embedding deep learning models into wearable systems (e.g. wristband or wristwatch), facilitating more responsive and efficient gesture recognition in daily-life applications. The Python implementation of the complete knowledge distillation framework is publicly available at: https://github.com/Open-EXG/handDistill.

LT-RNet: A Hybrid BILSTM-Transformer Network for sEMG Hand Gesture Recognition

An Adaptive Multi-Representation Fusion Framework for EMG Pattern Recognition in Transradial Amputees

Electromyography (EMG)-based pattern recognition is a key enabling technology for intuitive upper-limb prosthetic control; however, amputee EMG signals show substantial intersubject variability and nonstationary characteristics, which can limit the effectiveness of conventional fixed feature representations. This study proposes an Adaptive Multi-Representation Fusion Framework that combines an enhanced Adaptive Linear Series Decomposition Learner (Adaptive LSDL) with Time-Domain Power Spectral Descriptors (TDPSD) for amputee movement classification. The proposed Adaptive LSDL extends conventional decomposition by using adaptive threshold-region optimization and multitransform signal analysis, enabling subject-specific identification of the most discriminative decomposition characteristics while preserving low computational complexity. The complementary properties of TDPSD and Adaptive LSDL are then exploited through feature-level fusion, producing a unified representation that captures both local temporal-spectral complexity and adaptive structural signal information. Experimental evaluation was conducted using the Ninapro DB3 dataset, comprising 11 transradial amputees, and a MYO armband amputee dataset, comprising 6 transradial amputees. Subject-specific five-fold cross-validation was performed using a Linear Discriminant Analysis (LDA) classifier. Adaptive LSDL consistently improved performance relative to the original fixed-threshold LSDL representation, while fusion improved classification accuracy for 7 of 11 DB3 amputees and 5 of 6 MYO amputees. The largest fusion gains reached 4.39 and 5.60 percentage points for the DB3 and MYO cohorts, respectively. Furthermore, the proposed framework demonstrated competitive performance relative to a lightweight real-time convolutional neural network benchmark while requiring substantially lower computational complexity. These findings indicate that adaptive decomposition and representation-level information fusion provide an effective approach for exploiting subject-specific neuromuscular information while remaining suitable for computationally efficient prosthetic control applications.

Hand Gesture Recognition Based on Multi-Scale Attention Graph Convolutional Network

Advances in artificial intelligence have made hand gesture recognition an important human–computer interaction modality. Graph convolutional networks (GCNs) are widely used for skeleton-based hand gesture recognition, yet their performance can be limited by weak semantic topology modeling, underused feature channels, and shallow spatio-temporal fusion. We propose a Multi-scale Attention Graph Convolutional Network (MA-GCN) that combines three components within one skeleton framework: a hybrid topology that augments physiological connections with semantic priors; a Gaussian Multi-Scale Channel Attention (GMCA) module for coordinate denoising and adaptive channel weighting; and a Local-Global Fusion Module (LGFM) that combines local convolutional features with channel-wise global attention. Ablation studies quantify the independent and joint contributions of these components. MA-GCN obtains Top-1 accuracies of 97.50%/95.95% on SHREC’17 Track and 94.29%/92.86% on DHG14/28 for the 14-/28-class settings. In a SHREC’17 Track-to-FPHA pre-train-then-fine-tune evaluation, it reaches 94.09% Top-1 accuracy, providing preliminary evidence that the proposed framework maintains effectiveness under cross-dataset transfer.