色彩感知中的脑电信号多域特征选择算法研究

针对目前在不同色彩感知中的脑电信号识别方面的研究还不多见，本文提出采用随机森林算法对信号的时域特征和频域特征进行最优组合的方法对不同色彩感知中的脑电信号进行识别。首先采用小波变换，对脑电信号进行7层分解，提取脑电信号在delta、theta、alpha和beta节律频带上的小波能量，并结合脑电信号在时域上的统计量偏度和峰度组成特征向量。然后通过基于随机森林的特征选择算法提取最优的特征组合方案，删除冗余的特征量。使用自适应增强算法进行分类识别，识别的平均正确率可达到85.07%。该结果表明使用本文所提出的特征提取与选择方法用于不同色彩感知中的脑电信号识别上是可行的，并且能够取得较好的识别率。 

Research on Multi-domain Features Selection Algorithm of EEG Signals in Color Sensing

As the research on the identification of electroencephalograph (EEG) signals in color sensing is still rare, In this paper, the method of using the random forest algorithm to optimally combine the time domain features and frequency domain features of signals is proposed to identify the EEG signals in different color perceptions.. Firstly, the wavelet transform was used to decompose EEG signals by 7 layers, then extracted the average energy of EEG signals in delta, thelta, alpha and beta bands based on wavelet coefficients, and combined the statistic skewness and kurtosis to form feature vectors. In order to reduce computation load and delete redundant features, this paper proposed an feature selection algorithm based on random forest. And the adaptive boosting (AdaBoost) algorithm was adopted to classify the EEG signals. Results show that the average accuracy of the proposed algorithm can achieve 85.07%. And it proves that this algorithm proposed can classify the EEG signals based on different colors stimuli well, and get a better recognition rate.