基于特征融合和粒子群优化算法的运动想象脑电信号识别方法

由于运动想象脑电信号的信噪比较低，特征提取和特征选择比较困难，无法获得较高的分类准确率。针对上述问题，该文提取了时域、频域和空间域3个观察面的特征，并采用粒子群优化算法结合随机森林分类器来进行特征筛选。具体过程为，首先根据R2图来对信号进行带通滤波；其次，使用小波软阈值和得分共空间模式算法进行去噪和通道筛选；然后，通过3种算法提取时频域和空间域特征，待特征融合之后使用基于随机森林分类器的评价指标作为PSO的适应度函数，进行特征选择；最后，运用3种分类器以及集成分类器来验证效果。实验结果显示，通过特征融合以及特征选择可以去除冗余信息，保留有效信息，最终的分类正确率达到98.3%，为该技术在医疗康复等领域应用提供了新的方法。

Recognition of Electroencephalographic Signals in Motor Imaging Based on Feature Fusion and Particle Swarm Optimization

The signal-to-noise ratio of EEG signal is low, feature extraction and feature selection are difficult, and high classification accuracy cannot be obtained. To solve these problems, this paper extracts the features of time domain, frequency domain and space domain, and uses particle swarm optimization algorithm combined with random forest classifier to screen the features. The specific process is as follows: firstly, the signal is bandpass filtered according to the R2 graph; then, the wavelet soft thresholding and scoring common space pattern algorithm are used for denoising and channel filtering; furthermore, the time-frequency domain and space domain features are extracted through three algorithms, and the evaluation index based on the random forest classifier is used as the fitness function of particle swarm optimization (PSO) after feature fusion for feature selection; finally, three classifiers and integrated classifiers are used to verify the effect. The experimental results show that through feature fusion and feature selection, redundant information can be removed and effective information can be retained. The final classification accuracy is 98.3%, which provides a new method for the application of this technology in medical rehabilitation and other fields.