Imperial College London
Portrait Picture

DrBennyLo

Faculty of Medicine, Department of Metabolism, Digestion and Reproduction

Visiting Reader
 
 
 
//

Contact

 

+44 (0)20 7594 0806benny.lo Website

 
 
//

Location

 

Bessemer BuildingSouth Kensington Campus

//

Summary

 

Publications

Citation

BibTex format

@article{Ravi:2017:10.1109/JBHI.2016.2633287,
author = {Ravi, D and Wong, C and Lo, B and Yang, G},
doi = {10.1109/JBHI.2016.2633287},
journal = {IEEE Journal of Biomedical and Health Informatics},
pages = {56--64},
title = {A deep learning approach to on-node sensor data analytics for mobile or wearable devices},
url = {http://dx.doi.org/10.1109/JBHI.2016.2633287},
volume = {21},
year = {2017}
}
Download

RIS format (EndNote, RefMan)

TY  - JOUR
AB  - The increasing popularity of wearable devices inrecent years means that a diverse range of physiological and functionaldata can now be captured continuously for applicationsin sports, wellbeing, and healthcare. This wealth of informationrequires efficient methods of classification and analysis wheredeep learning is a promising technique for large-scale data analytics.Whilst deep learning has been successful in implementationsthat utilize high performance computing platforms, its use onlow-power wearable devices is limited by resource constraints.In this paper, we propose a deep learning methodology, whichcombines features learnt from inertial sensor data together withcomplementary information from a set of shallow features toenable accurate and real-time activity classification. The design ofthis combined method aims to overcome some of the limitationspresent in a typical deep learning framework where on-nodecomputation is required. To optimize the proposed method forreal-time on-node computation, spectral domain pre-processingis used before the data is passed onto the deep learning framework.The classification accuracy of our proposed deep learningapproach is evaluated against state-of-the-art methods using bothlaboratory and real world activity datasets. Our results show thevalidity of the approach on different human activity datasets,outperforming other methods, including the two methods usedwithin our combined pipeline. We also demonstrate that thecomputation times for the proposed method are consistent withthe constraints of real-time on-node processing on smartphonesand a wearable sensor platform.
AU  - Ravi,D
AU  - Wong,C
AU  - Lo,B
AU  - Yang,G
DO  - 10.1109/JBHI.2016.2633287
EP  - 64
PY  - 2017///
SN  - 2168-2208
SP  - 56
TI  - A deep learning approach to on-node sensor data analytics for mobile or wearable devices
T2  - IEEE Journal of Biomedical and Health Informatics
UR  - http://dx.doi.org/10.1109/JBHI.2016.2633287
UR  - http://hdl.handle.net/10044/1/42700
VL  - 21
ER  -
Download