利用自我映射組織圖進行雙重任務下分心之腦波反應辨識

Abstract

駕駛者分心已經證實是造成車禍發生的重大原因之一，因此若能及早偵測到駕駛者心理狀態的變化並給予適當地回饋機制是重要的。因此，本論文以自我映射組織圖(Self-Organizing Map, SOM)來分析、辨識人類的腦電波(Electroencephalogram, EEG)，探討駕車行為下之目標物時距（Stimulus Onset Asynchrony, SOA）影響人類分心效應之腦部反應變化，其中SOM是模擬人類大腦學系過程與學習後結果的類神經網路架構。本論文分析、辨識的腦電波，是經過去除雜訊及獨立成份分析(Independent Component Analysis, ICA)處理後的前額區以及運動感覺區這兩個腦部皮質收集到的EEG訊號，再經過降低維度、特徵擷取、去除基準、消除差異、標準化、以及平滑化等前處理步驟後才是完整的輸入資料。本實驗建構的自我映射組織圖大小為25*25，上述的資料當成輸入並設定兩階段學習。結果顯示學習後的自我映射圖是呈現二維圖形，經由觀察可以清楚地分辨單一任務與雙重任務的腦波資料，特別是單純開車和單純回答數學這兩個任務的腦波資料分別群聚在此映射圖的兩個角落，而雙重任務之腦波資料則群聚於映射圖中央，儘管有一些神經元是交錯坐落在此圖形中，但經過標示神經元此一步驟，每一個類別的辨識正確率皆超過百分之90。藉由此一研究發現雖然人類的行為表現經有統計檢定沒有顯著性的差異，但是在腦波反應上的確是存在細微的變化，而原本個體間差異相當大的腦電波訊號，經過消除差異這一個演算法處理後，可以大幅降低個體間訊號強弱的差異，並且完整保留處理不同任務時腦電波訊號的差異性。

Driver distraction is widely recognized as a leading cause of car accident. It is important to detect and determine the mental condition during driver distraction. In this study the self-organizing map (SOM) is adopted to recognize of the cross-session variability in EEG dynamics for dealing with dual task involving driving and answering simple math questions in the stimulus onset asynchrony (SOA) conditions. EEG signal from the frontal and the motor cortex are integrated to use as the input data. Each trial of the input data was processed with removal of baseline, feature extraction, and normalization. Then, the processed data was recognized by the SOM which constructed 25*25 maps through a two phase training scheme. Our results demonstrated that five cases (three dual-task and two single-task cases) can be distinguished clearly by the SOM-based method. Especially each single-task case was clustered in a distinct spatial area of the maps and the other dual-task cases showed several subgroups in the middle of the maps. Although some neurons were mixed in the maps, the accuracy of each case was higher than 90% after labeling. In conclusion, even if there was no significant difference in the behavioral data between two cases, such as response time and driving performance, the proposed SOM-based exploratory algorithm using EEG suggested existence of distinct signatures among the five cases. We have also suggested a method to reduce the variation among subjects for the same task and thereby could yield better maps.