TY - GEN
T1 - Human Activity Classification Using Recurrence Plot and Residual Network
AU - Lew, Ching Hong
AU - Lim, Kian Ming
AU - Lee, Chin Poo
AU - Lim, Jit Yan
N1 - Publisher Copyright:
© 2023 IEEE.
PY - 2023
Y1 - 2023
N2 - Human Activity Classification (HAC) is a challenging time series classification task, aiming to discern specific behaviors or movements of individuals based on sensor data. This paper proposed an innovative method for HAC that leverages the power of Residual Network and the unique insights provided by Recurrence Plots (RP). Recurrence Plots are a visualization technique that transforms time series data into graphical representations, highlighting temporal patterns and dependencies. In this proposed method, Recurrence Plots play a crucial role by capturing intricate temporal relationships in the sensor data, enabling more precise activity recognition. Furthermore, Residual Network (ResNet), a prominent deep learning architecture, is employed to create a robust and efficient classification model. ResNet is able to mitigate the vanishing gradient problem through residual connections, which is particularly advantageous in the context of time series classification. The experimental results on MotionSense, UCI-HAR, and USC-HAD datasets demonstrate the superiority of our proposed approach. When compared to five existing methods and a self-constructed transfer learning model, the proposed method consistently outperforms others, achieving the highest average accuracy of 84.01%. Notably, it reaches accuracy rates of 92.31%, 84.72%, and 75.00% for MotionSense, UCI-HAR, and USC-HAD datasets, respectively.
AB - Human Activity Classification (HAC) is a challenging time series classification task, aiming to discern specific behaviors or movements of individuals based on sensor data. This paper proposed an innovative method for HAC that leverages the power of Residual Network and the unique insights provided by Recurrence Plots (RP). Recurrence Plots are a visualization technique that transforms time series data into graphical representations, highlighting temporal patterns and dependencies. In this proposed method, Recurrence Plots play a crucial role by capturing intricate temporal relationships in the sensor data, enabling more precise activity recognition. Furthermore, Residual Network (ResNet), a prominent deep learning architecture, is employed to create a robust and efficient classification model. ResNet is able to mitigate the vanishing gradient problem through residual connections, which is particularly advantageous in the context of time series classification. The experimental results on MotionSense, UCI-HAR, and USC-HAD datasets demonstrate the superiority of our proposed approach. When compared to five existing methods and a self-constructed transfer learning model, the proposed method consistently outperforms others, achieving the highest average accuracy of 84.01%. Notably, it reaches accuracy rates of 92.31%, 84.72%, and 75.00% for MotionSense, UCI-HAR, and USC-HAD datasets, respectively.
KW - Human Activity Classification
KW - Recurrence Plot
KW - Residual Network
UR - http://www.scopus.com/inward/record.url?scp=85186719251&partnerID=8YFLogxK
U2 - 10.1109/ICSPC59664.2023.10420336
DO - 10.1109/ICSPC59664.2023.10420336
M3 - Conference Proceeding
AN - SCOPUS:85186719251
T3 - 2023 IEEE 11th Conference on Systems, Process and Control, ICSPC 2023 - Proceedings
SP - 78
EP - 83
BT - 2023 IEEE 11th Conference on Systems, Process and Control, ICSPC 2023 - Proceedings
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 11th IEEE Conference on Systems, Process and Control, ICSPC 2023
Y2 - 16 December 2023
ER -