A Deep Learning-based Model for Human Activity Recognition using Biosensors embedded into a Smart Knee Bandage
Sakorn Mekruksavanich, Ponnipa Jantawong, Anuchit Jitpattanakul
Abstract
Today, biosensors are effectively utilized in illness detection, prevention, rehabilitation, patient medical monitoring, and personal health promotion. The advancement of biosensing technology offers rigorous methods for measuring an individual's motor function, which contributes significantly to restoring motor function. An smart knee bandage was recently designed to analyze and decode diverse biosignals recorded by knee bandage-integrated biosensors. This research aims to examine human activity recognition (HAR) using biosignals gathered by a smart knee bandage. To accomplish this, a deep residual neural network was established for biosensor-based HAR. For performance assessment, we utilized a significant public biosensor dataset known as the CSL-SHARE dataset, which gathers electromyography (EMG), electrogoniometer (EGM), and inertial measurement unit (IMU) signals of individuals completing 22 simplex and complicated everyday tasks. Five-fold cross-validation was used to train and assess deep learning models. According to our findings, integrating features derived from the EMG, EGM, and IMU signal produced the maximum accuracy of 91.60% and the highest F1-score of 92.13%.