Lower limb motion recognition based on surface electromyography

Because the noise in lower limb surface electromyography (sEMG) is so high, extracting characteristics and recognizing lower limb motion is difficult. This research offers an improved threshold algorithm based on parameter a for sEMG signal denoising, convolution neural network (CNN) feature extraction, and kernel extreme learning machine (KELM) classification of lower limb motion recognition to address these issues. First, the sEMG signal decomposes into multiple variational mode functions (VMF) by variational mode decomposition, and the denoised sEMG signal is reconstructed after the VMF is processed by the enhanced threshold algorithm. The denoised sEMG signal is then sent into the CNN network as a matrix to achieve independent feature extraction of the sEMG signal. Finally, KELM is used to classify the retrieved feature signal. The experimental results reveal that this approach has a high signal-to-noise ratio, the root mean square is small, and a good denoising impact. The accuracy of lower limb classification is 95.90%, Based on traditional features and CNN, it is 18.29% and 11.86% higher than KELM, respectively. It can be seen that this research method can improve the accuracy of lower limb motion recognition.