Recent progress has led to the widespread use of Machine Learning (ML) in various fields, including biomedical applications. ML can assist in the diagnosis by detecting disease-specific patterns and monitoring the patient’s condition. Furthermore, ML algorithms can be implemented on resource-limited systems thanks to emerging technologies like the Internet of Things (IoT), Edge Computing, and wearable devices, enabling real-time disease detection and faster processing. This research explores wearable Neural Network (NN) applications using inertial signals. A multi-purpose wearable device has been developed for arm gesture recognition, fall detection, stair climbing and descending recognition, and gait and physical activity level monitoring. NN-based classifiers have been trained on the Edge Impulse platform and implemented on nRF52840 microcontroller. These classifiers achieved high accuracy (from 95.74% up to 100%), using a small memory portion (from 4.4kB to 9.4 kB of RAM and from 17.4kB to 99.8 kB of Flash memory).
Neural Network-based Wearable Devices for Limb Rehabilitation by Inertial Signal Classification
R. De FazioPrimo
;L. Spongano;P. Visconti
Ultimo
Writing – Original Draft Preparation
2024-01-01
Abstract
Recent progress has led to the widespread use of Machine Learning (ML) in various fields, including biomedical applications. ML can assist in the diagnosis by detecting disease-specific patterns and monitoring the patient’s condition. Furthermore, ML algorithms can be implemented on resource-limited systems thanks to emerging technologies like the Internet of Things (IoT), Edge Computing, and wearable devices, enabling real-time disease detection and faster processing. This research explores wearable Neural Network (NN) applications using inertial signals. A multi-purpose wearable device has been developed for arm gesture recognition, fall detection, stair climbing and descending recognition, and gait and physical activity level monitoring. NN-based classifiers have been trained on the Edge Impulse platform and implemented on nRF52840 microcontroller. These classifiers achieved high accuracy (from 95.74% up to 100%), using a small memory portion (from 4.4kB to 9.4 kB of RAM and from 17.4kB to 99.8 kB of Flash memory).I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
