基於低價位手環穿戴式裝置之睡眠辨別演算法

Sleep and Wake Discrimination with Low Price Wristband Device

劉守哲(Shou-Che Liu)

指導教授 ： 連豊力

國立臺灣大學/電機資訊學院/電機工程學研究所/碩士(2020年)

https://doi.org/10.6342/NTU202000562

摘要

睡眠不足日益被認為是健康的重要預後參數，並可能導致白天嗜睡。如果人們睡眠不足，可能會在運輸和工廠廠房中造成一些事故，並可能造成生命或財產損失。為避免這種情況，需要不斷警告人們“需要休息”。由於穿戴繁瑣的設備可能會影響人們的日常生活，因此許多人不願意將其戴在身上。本文提出了一種基於時間序列的睡眠檢測方法，該方法具有價格低，重量輕，舒適度高的可穿戴設備Garmin Vivosmart3。儘管低重量設備使人更舒適，但是它具有一個關鍵問題，即其接收生理信號的準確性較低。接收生理信號的準確性較低，可能會導致大量缺失值。本文提出的演算法可以通過減少資料量的方法來解決這個問題。而且由於睡眠是一種不能中斷的連續行為，因此本文中收集的數據被視為典型的時間序列數據集。針對時間序列數據集的特點，提出了一種基於連續序列發現的睡眠檢測方法。通過進行的幾次實際實驗，在所有受試者中報告的準確性，敏感性和特異性分別為94.79％，97.13％和93.86％。

關鍵字

穿戴式裝置 ； 缺失值處理 ； 時間序列 ； 睡眠檢測

並列摘要

Insufficient sleep is increasingly being recognized as an important prognostic parameter of health and could result in daytime sleepiness. If people do not have enough sleep, they may cause some accidents in transportation and factory plants and would loss their lives or properties. To avoid this circumstance, a constant warning about the people’s “need for rest” is required. Since wearing cumbersome equipments may affect people’s dayily lives, many peoples are not willing to put them on their bodies. This thesis proposes a time series based sleep detecting method with a low price, low weight and high comfort wearable device called Garmin Vivosmart3.
Although the low weight device is more comfortable for people, it has a critical problem which is its low accuracy of receiving physiological signals. Low accuracy of receiving physiological signals may cause large amount of missing values. In this thesis, the proposed algorithm could solve this problem with the aid of a data volume reduction method. And since sleep is a continuous behavior that cannot be interrupted, the collected data in this thesis is considered as a typical time series data set. Due to the characteristic of time series data set, a sleep detection method based on finding continuous sequences is proposed. With several real world experiments conducted, an accuracy, sensitivity, and specificity of 94.79%, 97.13%, and 93.86% is reported over all subjects.

並列關鍵字

Wearable device ； missing value processing ； time series ； sleep detection

參考文獻

[1: Walsh et al. 2017]

Lorcan Walsh, Seán McLoone, Joseph Ronda, Jeanne F. Duffy, and Charles A. Czeisler, “Noncontact Pressure-Based Sleep Wake Discrimination,” IEEE Transactions on Biomedical Engineering, Vol. 64, No. 8, pp. 1750-1760, Aug. 2017.

[2: Sano et al. 2019]

Akane Sano, Weixuan Chen, Daniel Lopez-Martinez, Sara Taylor, and Rosalind W. Picard, “Multimodal Ambulatory Sleep Detection Using LSTM Recurrent Neural Networks,” IEEE Journal of Biomedical and Health Informatics, Vol. 23, No. 4, pp. 1607-1617, July 2019.

[3: Hwang et al. 2017]