Recognition of Shoulder Exercise Activity Based on EfficientNet Using Smartwatch Inertial Sensors
Issued Date
2022-01-01
Resource Type
Scopus ID
2-s2.0-85141583099
Journal Title
2022 3rd International Conference on Big Data Analytics and Practices, IBDAP 2022
Start Page
6
End Page
10
Rights Holder(s)
SCOPUS
Bibliographic Citation
2022 3rd International Conference on Big Data Analytics and Practices, IBDAP 2022 (2022) , 6-10
Suggested Citation
Hnoohom N., Chotivatunyu P., Mekruksavanich S., Jitpattanakul A. Recognition of Shoulder Exercise Activity Based on EfficientNet Using Smartwatch Inertial Sensors. 2022 3rd International Conference on Big Data Analytics and Practices, IBDAP 2022 (2022) , 6-10. 10. doi:10.1109/IBDAP55587.2022.9907217 Retrieved from: https://repository.li.mahidol.ac.th/handle/20.500.14594/84349
Title
Recognition of Shoulder Exercise Activity Based on EfficientNet Using Smartwatch Inertial Sensors
Author's Affiliation
Other Contributor(s)
Abstract
Recognition of human activity is an important research topic due to its potential applications in areas, such as the medical industry and other related fields. Sensor-based human activity recognition (HAR) employing deep learning (DL) techniques has grown in popularity as a result of its immense efficiency in detecting complicated tasks and its low cost in comparison to more traditional machine learning (ML) techniques. More recently, convolutional neural network (CNN) approaches have also been used with sensor-based HAR, and the results have shown promise for improvement. In this study, we applied the CNN-based architecture known as EfficientNet to perform sensor-based HAR. The goal of this research was to apply the EfficientNet architecture to the classification of complicated human activities, such as shoulder workout activities. To evaluate the recognition performance, we used a benchmark HAR dataset called SPARS9x, which collected sensor data from six sports activities. According to the results of the experiments, the EfficientNet-B3 had the best performance overall on the benchmark dataset receiving an F1-score of 98.87%.