基于最优频段选择的运动想象研究

Research on Motor Imagery Based on Optimal Frequency Band Selection

脑电图(EEG)信号包含多个波段,其中运动想象主要涉及mu和beta波段。在明确所需频率的情况下,分类任务通常能够顺利执行。然而,频段选择并未成为主要研究焦点。因此,开发一种能够筛选出最优频段以参与分类任务的应用显得尤为重要。本研究基于BCI竞赛Ⅳ数据集2b,分析了视觉诱发的左右手运动想象脑电数据的分类效果。实验采用10折交叉验证方法对模型进行评估,旨在降低随机性并提供稳定性能指标。本研究应用了BP-CNN和AR-CNN两种模型,并与在已知频率范围内扩展频段的CNN模型进行了比较。即便在未知频率范围内,通过精确的频带选择和模型优化,仍能实现有效的运动想象识别,为脑机接口技术提供了理论基础,并对相关领域的未来研究具有重要的参考意义。

Electroencephalogram(EEG)signals consist of multiple frequency bands,with motor imagery primarily involving the mu and beta bands.When the required frequency bands are clearly defined,classification tasks can generally be performed smoothly.However,frequency band selection has not been a major focus of research.Thus,developing an application capable of screening the optimal frequency bands for classification tasks is particularly important.This research analyzes the classification performance of EEG data related to visual cue-based left-hand and right-hand motor imagery based on the BCI Competition IV dataset 2b.The experiments utilize a 10-fold cross-validation approach to evaluate the models,aiming to reduce randomness and provide stable performance metrics.This research applies two models of BP-CNN and AR-CNN,and compares them with a CNN model that extends the frequency bands within known ranges.Even within unknown frequency ranges,effective motor imagery recognition is achieved through precise frequency band selection and model optimization.It provides a theoretical foundation for brain-computer interface technology and offers significant reference value for future research in related fields.