A Lightweight Spatial Channel Reconstruction Network for sEMG Gesture Recognition

At present, gesture recognition models based on surface electromyography (sEMG) have difficulty achieving a balance between computational efficiency and high accuracy. To address this issue, we propose a spatial channel reconstruction convolutional neural network (SCR-CNN). The model comprises two primary units: the spatial reconstruct unit (SRU) and the channel reconstruct unit (CRU). The SRU uses a separation reconstruction methodology to effectively alleviate spatial redundancy, whereas the CRU adopts a separation transformation fusion strategy aimed at reducing channel redundancy and the computational cost. The sEMG signal is converted into a twodimensional image through the Gramian angular field and input into the SCR-CNN for feature extraction and classification. Evaluation across NinaPro dataset demonstrates that SCR-CNN exhibits strong performance in both computational efficiency and accuracy.