| Abstract: | 本研究透過人因實驗，模擬室內辦公場所，提取在不同環境音下使用者專注與放鬆狀態之腦波，經訊號處理及特徵萃取流程，將受試者不同情境的腦波數據使用多種分析工具進行腦波狀態辨識及評量，進一步探討不同聲音環境對使用者專注程度之影響，期許能幫助使用者注意自身專注力問題進而提升在工作、學習上的表現。
在環境噪音之專注度影響實驗中，受試者將會體驗兩輪的專注度遊戲測驗，每一輪對應的背景噪音與遊戲測驗內容不同，在接收實驗人員指示後，受試者應在不同聲音情境下進行專注度遊戲測驗或放鬆休息之動作，實驗過程中使用Neurosky耳機型腦波儀紀錄受試者即時的腦波訊號，作為專注度評估之數據，且受試者在專注度測驗中的得分表現、反應速度也會作為後續客觀分析的參考資料。
收集完受試者各情境間不同狀態之腦波數據，透過訊號處理的方式將雜訊干擾濾除，以減少雜訊干擾導致後續分析上的誤判，接著提取碎形維度(Fractal Dimension)、希爾伯特–黃轉換(Hilbert-Huang Transform, HHT)後得到的特定頻段之頻帶功率作為腦波分析的特徵指標，並利用接收者操作特徵曲線(Receiver operating characteristic curve, ROC Curve)、支持向量機(Support Vector Machine, SVM)及長短期記憶(Long Short-term Memory, LSTM)三種分析工具進行腦波狀態分類。
研究結果顯示去除雜訊對於使用碎形維度作為分析指標進行腦波狀態分類的效能有顯著的提升，特徵指標DSA、DSA3-100則在雜訊處理後分類效能微幅降低且大多無顯著差異。分類器的選擇上，使用ROC曲線之最佳閾值所決定的線性二元分類器需要的運算時間最短，其分類效能也為三者之中最低的;LSTM與SVM間的比較，在分類正確率上，LSTM較SVM更勝一籌，運算時間也較為省時。在專注度測驗表現的統計結果中發現，受試者不會受到聲音情境的影響，而是因練習效應而產生表現之差異。;In this study, we perform human factors experiments in a simulated office to extract the brainwaves of participants in their focused and relaxed states under different ambient sounds. Through signal processing and feature extraction methods, we evaluate and classify the brainwaves of participants in different situations by using a variety of analysis tools. By exploring the influence of different sound environments on users’ concentration, we hope to assist users in considering their own concentration status and improving their performance in work and study.
In the experiment of “the effect of ambient noise on concentration’’, participants will take two rounds of concentration games, and the tested sound and the content of the game corresponding to each round are different. After receiving the instructions from the experimenter, the participant will take a concentration game or relax in different sound situations. During the experiment, we use a Neurosky MindWave Mobile 2 to record the real-time brainwave data of participants for the evaluation of concentration status. The game scores and reaction time in the concentration game will also be used as reference materials for objective analysis.
The collected brainwave data are preprocessed to filter out the noise and interference, so as to reduce the misjudgment caused by the noise and interference in the succeeding analysis. Then, we extract the fractal dimension of the brainwave data and use the Hilbert-Huang transform (HHT) to extract the band powers of specific frequency bands as the features in the brainwave analysis. Furthermore, we use three analysis tools to classify the brainwaves in working and relaxing states, including receiver operating characteristic curve (ROC curve), support vector machine (SVM) and long short-term memory (LSTM).
The results show that the classification performance of brainwave states using fractal dimension is significantly improved after noise filtering. However, the alpha-band related indices “DSA” and “DSA3-100” show a slight decrease in classification accuracy after noise filtering, but no significant difference before and after the noise filtering in most cases. Among the three classifiers, the operation time for the linear binary classifier determined by the ROC curve is the shortest, but its classification accuracy is the lowest. The comparison between LSTM and SVM shows that LSTM is better than SVM in terms of classification accuracy, and the computation of LSTM is also more time-saving. From the scores and reaction time in the concentration games, the performance of the participants are not affected by the sound situations but by the practice effect. |
