基于频域信息的意识任务分类

Mental tasks classification based on frequency domain information

研究了频谱信息在意识任务分类中的应用。用傅里叶变换(FFT)和Burg方法来估计频谱信息,并对比了频谱信息和自回归模型系数在相同特征个数条件下对意识任务分类的作用。结果显示:①两种频谱估计方法取得了几乎相同的分类结果,频谱信息能提供更高的分类精度;②丢失高频信息会降低分类的效果;③对于较多特征个数的分类需要足够多的训练样本。FFT算法具有简单,速度快的特点,且使用多个特征量并不增加其计算量,这些优点使得傅里叶变换更适合于在线系统的应用与分析。

The use of power spectral information for mental tasks classification is studied. Both FFT and Burg algorithms are used to estimate power spectral information. The comparison of the effects of power spectral information and AR model coefficients on mental tasks classification is made under same condition. The results indicate： ① Both methods of FFT and Burg result in almost the same classification results, and greater classification accuracy can be achieved by using power spectral information; ② The loss of relatively high frequency band information decreases the accuracy of classification;③Much more feature vectors used for training are required by classification under the condition of there being many features. The method of FFT is simple and fast, and by using FFT algorithm computing more features has the same computationally demanding as computing one feature. All these advantages of FFT make it appropriate for online system application and analysis.