| DC 欄位 |
值 |
語言 |
| DC.contributor | 資訊工程學系 | zh_TW |
| DC.creator | 潘奕平 | zh_TW |
| DC.creator | Yi-Ping Pan | en_US |
| dc.date.accessioned | 2022-7-19T07:39:07Z | |
| dc.date.available | 2022-7-19T07:39:07Z | |
| dc.date.issued | 2022 | |
| dc.identifier.uri | http://ir.lib.ncu.edu.tw:88/thesis/view_etd.asp?URN=109522135 | |
| dc.contributor.department | 資訊工程學系 | zh_TW |
| DC.description | 國立中央大學 | zh_TW |
| DC.description | National Central University | en_US |
| dc.description.abstract | 疲勞駕駛在全球造成大量的傷亡。以臺灣為例，疲勞駕駛是佔所有意外的 20%。我 們定義了甚麼是疲勞，並且在相關研究中找到疲勞影響駕駛表現的證據。車道偏移量上 升、方向盤操控能力下降、反應時間變長、油門煞車控制（速度控制）變差，都是疲勞 駕駛會造成的。我們發現駕駛人本人很難認知道自己已經發生疲勞駕駛，因此需要透過 疲勞偵測系統輔助。我們分析了四種常見的疲勞駕駛系統及它們的優缺點，並且發現使 用 EEG 偵測疲勞是最精準且客觀的。可惜的是 EEG 目前無法使用在日常生活中，因為它使用非常麻煩且昂貴。我們的目標是能夠用把用降低腦電圖偵測疲勞的通道數，我們即可使腦電圖疲勞偵測落實在生活中。
本篇論文我們先是分析各種 EEG 分析方法、特徵選擇的方法。我們並接著研究篩 選器的架構: SVM 與 LSTM 的架構。我們的方法主要是盡可能的減少 40 個 EEG 通道 之間的關聯程度，並且利用篩選方法搭配上 LSTM 的機器分類方法，以達成減少通道 數。我們的結果，呼應我們的假設。在皮爾森的線性相關係數演算法中，我們使用 7 個 通道便達成 94.9 % 的疲勞準確度。而在相互資訊的演算法中，我們使用 6 個通道便達成 94.04 % 的疲勞準確度。
我們比較我們篩選出來的通道數，和當代 fMRI 觀察疲勞對於腦袋中的影響，兩者 皆顯示疲勞跟大腦的運動皮層有關。我們也和當代不同腦電圖特徵選擇應用方法比較， 並顯示我們的方法不論是在準確度還是減少的通道數都是比大多數的研究優秀。 | zh_TW |
| dc.description.abstract | Driver fatigue is an underestimated factor that leads to traffic accidents. This research reviewed researches and found out how drowsiness will impact driving performance in various ways. Increasing lane drifting, decreasing steering wheel controlling ability, increasing reaction time (RT), and poor speed control, are the signs and indicators for drivers to discover they are not suitable to operate the vehicle. We then reviewed the four most common ways to detect drowsiness. Various drowsiness detection methods this research reviewed, drowsiness detection via EEG has the highest accuracy. However, due to the multi-channel nature of the EEG signal, using EEG detection in everyday life nowadays is not possible. Our objective is to reduce the channel needed so that we could make EEG drowsiness detection useful in daily life.
First, this research looks into different ways to analyze EEG signal. This research inspects two feature selection methods, filter method and wrapper method. We also discussed two classification model, which is the widely used support vector machine (SVM) and the long short term model (LSTM). Our method introduced the feature selection method into drowsiness detection with the key idea of reducing the redundancies between 40 EEG channels with the combination of filter method and LSTM classifier. The result is that we obtain the accuracy of 94.9 % in 7 channel Pearson’s correlation method and 94.04 % in 6 channel mutual information method.
This research then evaluates the selected channel with some drowsiness research, which suggests that when it comes to drowsiness, this research should look into one’s motor cortex of the brain. This research compares our result to some state-of-the-art method in different EEG applications which show great results in both accuracy and the amount of channel reduced. | en_US |
| DC.subject | 疲勞偵測 | zh_TW |
| DC.subject | 特徵選擇 | zh_TW |
| DC.subject | 腦電圖 | zh_TW |
| DC.subject | 長短期記憶模型 | zh_TW |
| DC.subject | 遞歸神經網路 | zh_TW |
| DC.subject | 通道減少 | zh_TW |
| DC.subject | Feature Selection | en_US |
| DC.subject | EEG | en_US |
| DC.subject | fatigue detection | en_US |
| DC.subject | drowsiness detection | en_US |
| DC.subject | LSTM | en_US |
| DC.title | 用特徵選擇減少疲勞偵測腦電圖通道數 | zh_TW |
| dc.language.iso | zh-TW | zh-TW |
| DC.title | Channel Reduction for EEG Fatigue Detection Using Pearson’s Correlation and Mutual Information | en_US |
| DC.type | 博碩士論文 | zh_TW |
| DC.type | thesis | en_US |
| DC.publisher | National Central University | en_US |