| 摘要: | 近年來，穿戴式裝置已成為人們生活中，不可或缺的3C產品之一，幫助我們解決生活中許多問題，並為帶給我們日常生活更多便利，其中常見的應用為，使用行動裝置監控一個人行走的步數、心率及睡眠品質等，而睡眠是人生活中很重要的一環，大多數人都有睡眠障礙的問題，為能夠透過穿戴式裝置來幫助入眠，因此，本論文提出一種基於類神經網路的睡眠效率評估系統。
此系統是透過多層感知機和放射狀基底函數網路的模型，並利用穿戴式裝置所收集使用者的資料預測睡眠效率，讓使用者可以預期今天睡眠狀況，其中，收集的資料包含使用者基本資料、心率、卡路里、步伐，以及飲食等與使用者睡眠相關資料，此外，本論文亦針對收集資料的特徵，且使用皮爾森係數、主成份分析及費雪係數，進一步分別檢驗每個特徵對於睡眠效率的相關度，此外，更進一步於系統中增加氣氛裝置之功能，讓使用者能夠利用此裝置調整環境，以提升睡眠效率。
在本論文的實驗中，我們比較現有的睡眠資料集和自己收集的資料集的差異，亦加入深度學習的深度神經網路，並針對各種類神經網路架構比較辨識效果。
;In recent years, wearable devices are one of indispensable products. The advantages of wearable devices that help us to solve the problem and bring us a more convenient daily life, such as use of wearable devices to monitor a person walking steps, heart rate and the quality of sleep, etc. Nowadays, Sleep issue has become increasing seriousness, there are more and more people suffer from insomnia or sleep disorder. In order to assess sleep efficiency by wearable devices. Therefore, this thesis is about to proposes a sleep efficiency assessment system based on artificial neural networks.
This system which based on MLP and RBFN, including users’ basic information, heart rate, calories, step, diet and sleep-related information to predict sleep efficiency, which also uses the Pearson coefficient, principal component analysis and Fisher ratio respectively, for the test feature to collect data for each feature correlation sleep efficiency, an additional function of the atmosphere device in the system, allowing users to take advantage of this means for adjusting environment, improve sleep efficiency.
The experiment in this thesis has compared to other sleep dataset and our dataset. We also refer to deep neural network to test the datasets, and compare accuracy for artificial neural networks. |
