A study on light weight of radar-based human activity recognition model using self-attention mechanism
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
Recently, radar-based human activity recognition (HAR) technology has been actively studied in the field of artificial intel-ligence by applying radar datasets to deep learning (DL) models to automatically learn and classification. This is one of the important applications in the field of activity recognition and can be used in various fields, such as exercise tracking, smart homes, self-driving cars, and health status monitoring, by recognizing human daily activity patterns. However, DL models are complex and have significant computational costs and numerous parameters to process and classify high-dimensional radar datasets. Therefore, their implementation on commercial mobile devices is limited by their computational complexity. Therefore, in this study, we propose a lightweight HAR DL model using Self-Attention technology to solve the complexity and computational costs of these DL models. Experimental results demonstrate that this model can maintain its performance while reducing the number of parameters required. In the future, these light-weight models will not only be usable on mobile devices but will also require lower computing power and memory capacity; therefore, they are expected to be used in various fields as time series-based DL models.
Abstract
Recently, radar-based human activity recognition (HAR) technology has been actively studied in the field of artificial intel-ligence by applying radar datasets to deep learning (DL) models to automatically learn and classification. This is one of the important applications in the field of activity recognition and can be used in various fields, such as exercise tracking, smart homes, self-driving cars, and health status monitoring, by recognizing human daily activity patterns. However, DL models are complex and have significant computational costs and numerous parameters to process and classify high-dimensional radar datasets. Therefore, their implementation on commercial mobile devices is limited by their computational complexity. Therefore, in this study, we propose a lightweight HAR DL model using Self-Attention technology to solve the complexity and computational costs of these DL models. Experimental results demonstrate that this model can maintain its performance while reducing the number of parameters required. In the future, these light-weight models will not only be usable on mobile devices but will also require lower computing power and memory capacity; therefore, they are expected to be used in various fields as time series-based DL models.
참고문헌 (16)
- H. F. Nweke, "Data fusion and multiple classifier systems for human ac-tivity detection and health monitoring: Review and open re-search directions", Information Fusion, 46, pp.147~170 (2019)
- A. B. Mabrouk, "Abnormal behavior recognition for intelligent video surveillance systems: A re-view", Expert Systems with Applications, 91, pp.480~491 (2018)
- M. Z. Uddin, "Human activity recognition us-ing wearable sensors, discriminant analysis, and long short-term memory-based neural structured learning", Scientific Reports, 11 (2021)
- J. Le Kernec, "Radar signal processing for sensing in assisted living: The challenges associated with real-time im-plementation of emerging algorithms", IEEE Signal Pro-cessing Magazine, 36(4), pp.29~41 (2019)
- Y. Kim, "Human detection and activity classi-fication based on micro-Doppler signatures using deep con-volutional neural networks", IEEE Geoscience and Remote Sensing Letters, 13(1), pp.8~12 (2016)
- Activity recogni-tion of FMCW radar human signatures using tower convo-lutional neural networks, "A. Helen Victoria", Wireless Networks, 1~17 (2021)
- F. J. Abdu, "Activity classification based on feature fusion of FMCW radar human motion mi-cro-Doppler signatures", IEEE Sensors Journal, 22(9), pp.8648~8662 (2022)
- X. Bai, "Radar-based human gait recognition using dual-channel deep convolutional neu-ral network", IEEE Transactions on Geoscience and Remote Sensing, 57(12), pp.9767~9778 (2019)
- J. Zhu, "A hybrid CNN–LSTM net-work for the classification of human activities based on mi-cro-Doppler radar", IEEE Access, 8, pp.24713~24720 (2020)
- H. -U. -R. Khalid, "Multi-view CNN-LSTM architecture for radar-based human activity recognition", IEEE Access, 10, pp.24509~24519 (2022)
- S. Zhu, "Con-tinuous human activity recognition with distributed radar sensor networks and CNN–RNN architectures", IEEE Transactions on Geoscience and Remote Sensing, 60, pp.1~15 (2022)
- W. Ding, "Radar-based human ac-tivity recognition using hybrid neural network model with multidomain fusion", IEEE Transactions on Aerospace and Electronic Systems, 57(5), pp.2889~2898 (2021)
- X. Li, "Human motion recognition with limited radar micro-Doppler signatures", IEEE Transactions on Geosci-ence and Remote Sensing, 59(8), pp.6586~6599 (2021)
- Z. Cao, "Towards cross-envi-ronment human activity recognition based on radar without source data", IEEE Transactions on Vehicular Technology, 70(11), pp.11843~11854 (2021)
- MobileNets: Efficient convolutional neural networks for mobile vision applications, "A. G. Howard", 1~9 (2017)
- X. Zhang, "ShuffleNet: An extremely efficient convolutional neural network for mobile devices", Proceedings of the IEEE conference on computer vision and pattern recognition, pp.6848~6856 (2018)