Title: Improving activity recognition for multiple-node wireless sensor network system based on compressed sensing
Authors: Duo Yang; Jiangtao Huangfu
Addresses: Laboratory of Applied Research on Electromagnetics, Zhejiang University, Hangzhou 310027, China ' Laboratory of Applied Research on Electromagnetics, Zhejiang University, Hangzhou 310027, China
Abstract: The paper proposes a wireless sensor network activity recognition system based on compressed sensing (CS) for data compression-transmission-recovery (CTR-WSN) in the case of data loss, where effective use of transmission resources, real-time performance, low data recovery errors, and high activity recognition accuracy are achieved. It performs different degrees of compression according to types of activities and transmission resources are effectively used. The most time-saving data recovery methods are evaluated based on SP/SAMP, and the error of them are basically controlled within ±0.03 when using 5000 samples. The recognition accuracy is improved by 10%, where SVM gets the highest result, reaching 94.09%. With the increase of sensor nodes and the data loss, the accuracy can maintain at a higher level, which is 36% higher than the original data when testing 24 nodes. This system will have broad application prospects in the fields of education and medical care in the future.
Keywords: activity recognition; compressed sensing; data loss; multiple-node; wireless sensor network; compression-transmission-recovery; acceleration sensor; machine learning; data recovery; real-time.
International Journal of Sensor Networks, 2020 Vol.34 No.3, pp.162 - 171
Received: 09 Nov 2019
Accepted: 29 Apr 2020
Published online: 26 Oct 2020 *