Feature Selection and Reduction of Lower Limb Activity Recognition Based on Surface Electromyography and Motion Data

Abstract

Daily activity recognition of lower limbs is of great significance to the health care of the elderly and patients with hemiplegia. Surface electromyography (sEMG) signal can directly reflect neuromuscular activity and is an important method for non-invasive monitoring of muscle activity on the body surface. In this paper, a novel method based on sEMG signal and inertial measurement unit (IMU) data to recognize daily activities of lower limbs is proposed. Record sEMG signals and IMU data of fifteen subjects using wearable sensor devices. After preprocessing such as filtering and sliding windows on the data, we extracted seventeen features. A feature selection method based on maximal relevance and minimal redundancy maximal relevance (mRMR) to select representative features. The selected features are input into four machine learning classifiers to classify four daily activities. The performance of the classifier is evaluated using accuracy and receiver operating characteristic curve-area under curve (ROC-AUC) score. The results show that the support vector machine has excellent performance in recognizing the daily activities of human lower limbs.