Multimodal Inertial Sensing Signal Rehabilitation Action Recognition Method Based on GAF-MultiRocket

DOI: https://doi.org/10.62517/jmhs.202605314

Author(s)

Lei Hua

Affiliation(s)

Department of Sports Rehabilitation, Shanxi Medical University, Taiyuan, China

Abstract

Aiming at the task of wearable inertial sensing rehabilitation action recognition in the home rehabilitation scenario, to fully exploit the motion features of multi-node and multimodal inertial time-series signals and reduce the recognition error caused by the posture perturbation of sensor wearing, this paper proposes the Grouped Awareness Multi-View Fusion MultiRocket (GAF-MultiRocket) algorithm. This algorithm constructs three complementary feature branches: the global view, the grouped awareness view, and the amplitude view, extracts multi-scale time-series features using random convolution and various pooling operators, constructs rotation-invariant amplitude features through the L2 norm, and completes classification using the closed-form solution of ridge regression after feature concatenation. Experimental verification is carried out based on the UCI Physical Rehabilitation Exercises public dataset. The experimental results show that the algorithm achieves an identification accuracy of 95.24% and a macro-average F1 score of 0.9480. The model does not require backpropagation iterative training and can quickly complete operations in the CPU environment, making it suitable for deployment on low-computing-power home rehabilitation devices, providing algorithm support for wearable intelligent rehabilitation action monitoring.

Keywords

Rehabilitation Action Recognition; Inertial Measurement Unit; Time Series Classification; Multirocket; Wearable Sensor

References

[1]Ettefagh A, Fekr A R. Enhancing automated lower limb rehabilitation exercise task recognition through multisensor data fusion in telerehabilitation[J]. Biomedical Engineering Online, 2024, 23(1):35. [2]Yi Chunzhi, Jia Yicheng, Jiang Feng, Wang Xiulai. Human Motion Intention Recognition Methods Based on Inertial Measurement Units: Status and Challenges[J]. CAAI Transactions on Intelligent Systems, 2025, 20(4): 763775. [3]Bulling A, Blanke U, Schiele B. A tutorial on human activity recognition using bodyworn inertial sensors[J]. ACM Computing Surveys, 2014,46(3):133. [4]Ronao C A, Cho S B. Human activity recognition with smartphone sensors using deep learning neural networks[J]. Expert Systems with Applications, 2016,59:235244. [5]Ojetola O, Gaura E, Brusey J. Shoulder physiotherapy exercise recognition: machine learning the inertial signals from a smartwatch[J]. Sensors, 2018,18(9):2875. [6]Yurtman A, Barshan B. Automated evaluation of physical therapy exercises using multitemplate dynamic time warping on wearable sensor signals[J]. Computer Methods and Programs in Biomedicine, 2014, 117(2): 189207. [7]Liang W, Wang F, Fan A, et al. Extended application of inertial measurement units in biomechanics: from activity recognition to force estimation[J]. Sensors, 2023, 23(9):4229. [8]Tan C W, Dempster A, Bergmeir C, et al. MultiRocket: multiple pooling operators and transformations for fast and effective time series classification[J]. Data Mining and Knowledge Discovery, 2022, 36(5):16231646. [9]Grandini M, Bagli E, Visani G. Metrics for multiclass classification: an overview[EB/OL].(20200812)[20260826]. [10]Du Yanchen, Yang Liangdong, Wang Xiaoming, Li Linrong, Liao Wenxiang, Yu Hongliu. Lowerlimb Motion Intention Recognition Based on Myopressure Signal and Inertial Measurement Unit[J]. Information and Control, 2024, 53(6): 761773. [11]Hadjidj A, Bouabdallah A, Challal Y, Souil M, Owen H. Wireless sensor networks for rehabilitation applications: challenges and opportunities[J]. Journal of Network and ComputerDempster A, Schmidt D F, Webb G I. MiniRocket: a very fast (almost) deterministic transform for time series classification[C]//Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining. New York: ACM, 2021:248257. [12]Dempster A, Schmidt D F, Webb G I. MiniRocket: a very fast (almost) deterministic transform for time series classification[C]//Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining. New York: ACM, 2021:248257.