基于FCA-ReliefF的融合生理信号情绪识别研究

Emotion Recognition Based on Physiological Signal Fusion and FCA-ReleifF

摘要:

针对当前情绪识别研究中特征维数多、识别率不高的问题,提出了基于多生理信号(心电、肌电、呼吸、皮肤电)融合及FCA-ReliefF特征选择的情绪识别方法。通过将从时域和频域两个维度提取的生理信号特征进行融合,作为分类器的输入进行情绪分类。为了降低特征维度,首先进行特征相关性分析(FCA)删除相关性较大的特征；再通过ReliefF剔除分类贡献弱的特征,达到降低特征维度的目的。在公开的数据集上进行验证,并与相关研究进行对比。结果表明,提出的方法在特征维度及识别率两个方面均有优势。提出的FCA-ReliefF降维策略有效地将特征从108维减少到60维,并且将识别精度提高到98.40%,验证了方法的有效性。

Abstract:

Focused on the problems of large feature dimension and low recognition rate in current emotional recognition research, an emotional recognition method based on feature fusion of multiple physiological signals (ECG, EMG, RSP, SC) and FCA-ReliefF algorithm is proposed. The features extracted from time domain and frequency domain are fused as the input of the classifier. In order to reduce feature dimension, feature correlation analysis (FCA) was used to eliminate features with strong correlation. Then ReliefF was used to delete features with weak classification contribution. Experimental analysis on the public dataset and compared with related studies show that the proposed FCA-ReliefF dimensionality reduction strategy can effectively reduce the feature dimension from 108 to 60, and the emotion recognition rate is improved up to 98.40%, which is better than the reported experimental results from the perspective of feature dimension and recognition accuracy.