.png){kind=icon}
.png){kind=icon}
Return
Wearable Sensor-based physical activity intensity recognition using deep learning feature engineering fusion
.png){kind=icon}
PRE
AI
DOI：10.1016/j.measurement.2024.115663.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Wearable devices have been widely utilized in areas such as rehabilitation medicine. These devices rely on algorithmic models to detect and provide feedback on the intensity of physical activity. However, a large amount of redundant data exists in the data acquisition process, resulting in poor model robustness. This study proposed an algorithm that combined long short-term memory and convolutional neural networks to extract features and utilized the deep subdomain adaptation network algorithm to construct a classification model. The average classification accuracy across the eight body parts reached 91.62%, representing an improvement of 12.23% compared with models without feature extraction. This study demonstrated the effectiveness of using domain adaptation approaches to address individual differences and the potential of long short-term memory coupled with convolutional neural networks for deep learning feature engineering. The results laid a theoretical foundation for future research using inexpensive sensors for measuring physical activity intensity across individuals.
Keywords:
Intensity recognition
Deep learning
Deep subdomain adaptation network
Feature extraction
Journal
IF:
5.6
Papers:
2.0W
Citations:
5.4W