基于多生理信息迁移学习的脑力负荷分类

在复杂的人机系统中,保持对实验人员脑力负荷状态的监测对于维护人机系统的安全、高效运行具有极为重要的理论和应用意义。针对现有脑力负荷分类方法识别率低及实际应用时测试样本数据偏移问题,本研究提出采用迁移学习及脑电和心电特征融合的脑力负荷分类识别方法,基于多任务航空情境操作的MATB-II平台同步采集12名健康受试者的脑电信号和心电信号,分别从时域和频域上提取各生理信息特征并进行融合,在此基础上引入迁移学习,基于迁移成分分析(Transfer Component Analysis,TCA)方法进行特征空间变换,实现源域和目标域的边缘分布适配,并进行脑力负荷分类。实验结果表明,基于多生理信息特征融合识别率高于传统脑力负荷识别方法,使用迁移学习可取得较高的识别准确率,为多生理信息脑力负荷分类研究提供了新方法。

Classification of Mental Workload Based on Multi-physiological Information Transfer Learning

Keep Monitoring mental workload state of personnel has extremely important theoretical and practical value for maintaining the safety and efficiency of man-machine system. A mental workload classification method based on transfer learning and features fusion of electroencephalogram(EEG) and electrocardiograph (ECG) was proposed aiming at the low recognition rate and the test sample data migration of the existing mental workload classification methods. The EEG and ECG signals of 12 healthy subjects were collected synchronously on the MATB-II platform for multi-task aviation situation operation, and the physiological information features were extracted and fused, from time domain and frequency domain respectively. On this basis, transfer learning was introduced. Based on Transfer Component Analysis (TCA) method, the feature data of the source domain and target domain was transferred to the common low-dimensional subspace, in which the edge distribution of source domain and target domain was adapted. The recognition rate of multi-physiological information feature fusion is higher than that of traditional mental workload identification methods, and transfer learning can achieve a higher recognition accuracy based on the experiment results, which provides a new method for multi-physiological information mental workload classification research.