EEG信号的时间,频率和空间信息对于运动图像解码至关重要(Zheng等,2022)。因此,已经广泛研究了基于时间频率空间特征的运动图像解码(Chen等,2023)。In the process of temporal- frequency-spatial feature extraction, the original EEG signals are first decomposed into multiple time-frequency units, then the common spatial pattern (CSP) algorithm is used to extract the spatial features on each time-frequency unit, and finally, the spatial features of multiple time-frequency units are cascaded into a feature vector ( Miao et al., 2021 ), which significantly increases脑电图的特征维度。特征尺寸的数量超过一百甚至几百,而特征冗余。此外,由于收集脑电图样本的困难和高成本,尤其是对于患者,样本量通常相对较小,通常只有几十个。因此,时间频率空间特征是高维小样本数据,它将为EEG分类模型带来一系列问题,例如过度拟合和模型解决方案的问题不确定(Chadebec等,2022)。