东北大学学报(自然科学版) ›› 2024, Vol. 45 ›› Issue (7): 928-935.DOI: 10.12068/j.issn.1005-3026.2024.07.003

• 信息与控制 • 上一篇    下一篇

基于多维特征矩阵和改进稠密连接网络的情感分类

李红利1(), 刘浩雨1, 张荣华2, 成怡1   

  1. 1.天津工业大学 控制科学与工程学院,天津 300387
    2.天津工业大学 人工智能学院,天津 300387
  • 收稿日期:2023-03-05 出版日期:2024-07-15 发布日期:2024-10-29
  • 通讯作者: 李红利
  • 基金资助:
    国家自然科学基金资助项目(62071328)

Emotional Classification Based on Multidimensional Feature Matrix and Improved Dense Connection Network

Hong-li LI1(), Hao-yu LIU1, Rong-hua ZHANG2, Yi CHENG1   

  1. 1.School of Control Science and Engineering,Tiangong University,Tianjin 300387,China
    2.School of Artificial Intelligence,Tiangong University,Tianjin 300387,China.
  • Received:2023-03-05 Online:2024-07-15 Published:2024-10-29
  • Contact: Hong-li LI
  • About author:LI Hong-liE-mail:lihongliln@163.com

摘要:

情感脑电信号是一种低信噪比的非平稳时间序列,传统的特征提取与分类方法难以提取不同情感状态时的有效特征并进行分类.针对以上情况,提出一种自动融合脑电信号不同频段和时频域特征的深度学习模型.首先,对预处理后的数据进行分频段处理,提取各频段的微分熵特征.然后,在网络中接入的挤压激励模块,对不同频段特征的微分熵特征赋权值,来获取输入数据的有价值信息,再利用改进的稠密连接网络进行特征融合并分类识别,保证了网络层之间最大程度的信息传输.最后,利用SEED情感脑电信号三分类数据集对算法进行了验证,分类正确率可达96.03%,高于现有的基线学习算法,所提算法可进一步增强网络特征提取能力,具有较快的收敛速度,对提升分类器的性能研究具有重要意义.

关键词: 情感分类, 稠密连接, 多维特征矩阵, 深度学习, 挤压激励

Abstract:

Emotional EEG(electroencephalogram) signal is a non?stationary time series with low signal?to?noise ratio. Traditional feature extraction and classification methods are difficult to extract and classify the effective features of different emotional states. In regard to the above situation, a deep learning model that automatically fuses different frequency bands and time?frequency characteristics of EEG signals is proposed. Firstly, the preprocessed data is processed in frequency bands, and the differential entropy features of each frequency band are extracted. Then, the squeeze excitation module connected in the network assigns weight to the differential entropy features of different frequency bands to obtain the valuable information of the input data, and then uses the improved dense connection network for feature fusion and classification recognition to ensure the maximum information transmission between the network layers. Finally, the algorithm is verified by using the SEED emotional EEG of three classification dataset, and the classification accuracy is 96.03%, which is higher than the existing baseline learning algorithm. The proposed algorithm further enhances network feature extraction capabilities and demonstrates faster convergence, which is of great significance for improving the performance of the classifier.

Key words: emotional classification, dense connection, multidimensional feature matrix, deep learning, squeeze excitation

中图分类号: