中国生物医学工程学报
中國生物醫學工程學報
중국생물의학공정학보
CHINESE JOURNAL OF BIOMEDICAL ENGINEERING
2009年
6期
851-855
,共5页
脑机接口%P300%滤波器%支持向量机
腦機接口%P300%濾波器%支持嚮量機
뇌궤접구%P300%려파기%지지향량궤
brain computer interface%P300%filter%support vector machine
基于P300电位的脑机接口是脑机接口系统研究中的重要范式之一.本研究探讨双极导联选择、滤波器设计和特征窗长度这三者中,参数选择对分类正确率的影响.使用的数据为6名受试者进行12选择Oddball范式的实验数据,数据分类采用支持向量机.结果表明,通过导联优选,对不同的受试者进行定制,可以有效提高分类正确率.滤波器频带对分类正确率有一定影响,但是滤波器阶次对结果影响不大.随着特征窗长度的增加,分类正确率也会提高,因此使用较长特征窗有助于提高分类正确率.本研究结果推荐的数据处理过程为:通过导联优选为每位受试者定制最优的双极导联;滤波过程使用0.5~16 Hz,阶次为20的FIR滤波器,然后把数据降采样为32 Hz;特征窗提取时可选择刺激后0~600 ms的数据.
基于P300電位的腦機接口是腦機接口繫統研究中的重要範式之一.本研究探討雙極導聯選擇、濾波器設計和特徵窗長度這三者中,參數選擇對分類正確率的影響.使用的數據為6名受試者進行12選擇Oddball範式的實驗數據,數據分類採用支持嚮量機.結果錶明,通過導聯優選,對不同的受試者進行定製,可以有效提高分類正確率.濾波器頻帶對分類正確率有一定影響,但是濾波器階次對結果影響不大.隨著特徵窗長度的增加,分類正確率也會提高,因此使用較長特徵窗有助于提高分類正確率.本研究結果推薦的數據處理過程為:通過導聯優選為每位受試者定製最優的雙極導聯;濾波過程使用0.5~16 Hz,階次為20的FIR濾波器,然後把數據降採樣為32 Hz;特徵窗提取時可選擇刺激後0~600 ms的數據.
기우P300전위적뇌궤접구시뇌궤접구계통연구중적중요범식지일.본연구탐토쌍겁도련선택、려파기설계화특정창장도저삼자중,삼수선택대분류정학솔적영향.사용적수거위6명수시자진행12선택Oddball범식적실험수거,수거분류채용지지향량궤.결과표명,통과도련우선,대불동적수시자진행정제,가이유효제고분류정학솔.려파기빈대대분류정학솔유일정영향,단시려파기계차대결과영향불대.수착특정창장도적증가,분류정학솔야회제고,인차사용교장특정창유조우제고분류정학솔.본연구결과추천적수거처리과정위:통과도련우선위매위수시자정제최우적쌍겁도련;려파과정사용0.5~16 Hz,계차위20적FIR려파기,연후파수거강채양위32 Hz;특정창제취시가선택자격후0~600 ms적수거.
The P300 based brain computer interface (BCI)is one of the important models in BCI study.This paper studied the parameter optimization in bipolar lead selection, filter design and temporal window length selection.The data came from 6 subjects in an oddball paradigm with 12 choices.Support vector machine was adopted for classification.The results illustrated that the classification accuracy could be enhanced by bipolar lead optimization for individuals.The pass band of the filter influenced on the classification accuracy, but the order of filter did not.Also, the classification accuracy could benefit from a long temporal window.With the above results, a procedure of parameter optimization was proposed.Bipolar lead was acquired for individuals during the data preprocessing at first.The data could be down-sampled to 32 Hz after a 0.5~60 Hz band pass FIR filter with an order of 20.A fixed temporal window of 0~600 ms after stimuli was a good choice for feature extraction.