Wireless body area sensor networks based human activity recognition using deep learning.

Area of Science:

• Biomedical Engineering
• Computer Science
• Artificial Intelligence

Background:

• Wireless Body Area Networks (WBANs) are crucial for remote patient monitoring, collecting health status and activity data.
• Sensor-based Human Activity Recognition (HAR) is gaining traction due to its privacy and convenience, amplified by IoT and wearable tech.
• Deep learning excels at automatic feature extraction, but challenges like environmental factors affect traditional computer vision methods.

Purpose of the Study:

• To propose and develop an advanced HAR system for WBANs.
• To leverage Gramian Angular Field (GAF) and DenseNet for improved HAR accuracy.
• To address limitations of existing HAR methods in real-world WBAN scenarios.

Main Methods:

• Time-series sensor data from WBANs was pre-processed to remove artifacts and apply median filtering.
• The Gramian Angular Field (GAF) algorithm transformed time-series data into 2D images.
• DenseNet was employed for automatic feature extraction and integration of multi-sensor data.

Main Results:

• The proposed GAF and DenseNet-based HAR system demonstrated superior performance.
• Achieved an accuracy of 97.83%.
• Recorded an F-measure of 97.83% and a Matthews Correlation Coefficient (MCC) of 97.64%.

Conclusions:

• The developed HAR system effectively utilizes GAF and DenseNet for accurate patient activity recognition in WBANs.
• This method offers a robust solution for healthcare monitoring, overcoming common challenges.
• The high accuracy indicates significant potential for clinical applications and enhanced patient care.