近年來，穩態視覺誘發電位(steady state visual evoked potential, SSVEP)為基礎之大腦人機界面(BCI)已被廣泛使用，藉由利用不同頻率對閃光進行進行編碼，系統可以經過分析穩態視覺誘發電位的頻率，辨別出使用者正在注視的閃光，並輸出對應的控制指令。相較於其它腦波人機介面，以穩態視覺誘發電位為基礎之腦波人機介面，具有高傳輸率(>20位元/分)與高準確率(>90%)，故受到研究學者的重視。然而，此種腦波人機介面在使用者移動視線注視不同選項時，會造成視覺誘發電位訊號的不明確而導致系統無法判別或誤判。為了解決此問題，本研究提出用匹配濾波器擷取訊號特徵，再利用隱馬可夫模型(Hidden Markov Model,HMM)，對誘發的腦波訊號建立模型，並藉由維特比解碼(Viterbi Decoding)找出隸屬各模型之最大機率以進行分類，達到提高訊號辨識度的效果。目前有3位受測者，平均正確率可達90.21%，平均ITR為32.9 bits/min。本研究使用隱馬可夫模型可針對不同選項間之轉換情形進行辨識，研究成果可提高視覺誘發腦波人機介面之穩定度，將來預計可進一步提升辨識速度，應用於遙控車之即時控制上。In recent years, steady-state visual evoked potential (SSVEP) – based brain computer interface (BCI) has been widely used in many applications. By tagging flickers with different frequencies, user’s gazed targets can be recognized by analyzing the frequencies of SSVEPs. SSVEP-based BCI has drawn great attentions by scientist and engineers due to its high information transfer rate (>20bits/min) and high accuracy (>90%). However, uncertainty and ambiguity usually occurs while user shifting his gaze between different targets, which usually result in incapability or, even worse, discrimination error in gazed target identification. Therefore, this study attempts to adopt hidden markov model (HMM) to classify the measured SSVEP into gazed, target-shifted and unattended states. The frequency contents in SSVEP were first analyzed by a match-filter detector and viterbi decoding was subsequently used to evaluate the probability of categories in HMM. The model which had maximum probability was recognized to activate its corresponding command. Currently, we have tested the proposed system on three volunteers with a mean accuracy of 90.21%. The experiment results validated the efficacy of HMM in discerning the transition states when subject shifting their gazes among different targets, which might be helpful to increase the stability of a SSVEP-based BCI. Future work will apply this system to the real-time control of a remote-control car.
