A CNN-LSTM Hybrid Model for Ankle Joint Motion Recognition Method Based on sEMG
Haoran Cheng, Guang‐Zhong Cao, Caihong Li, Aibin Zhu, Xiaodong Zhang
Abstract
A CNN-LSTM hybrid model for ankle joint motion recognition based on surface electromyography (sEMG) signals is proposed in this paper. The traditional recognition method is to manually extract the features from sEMG signals and then use machine learning method to train the model, which relies on prior knowledge and requires a lot of time to test and select good features to obtain high classification accuracy. In this paper, the CNN-LSTM hybrid model is used to identify four ankle joint movements (dorsiflexion, plantar flexion, foot varus and foot eversion). The hybrid model consists of two CNN layers and three LSTM layers. CNN can learn to automatically extract features and LSTM is able to capture long-term correlations of sEMG data. The experiment results show that the proposed model is effective and accurate, thus providing a basis for the subsequent research.