基于迁移学习Boosting的跨时间脑力负荷识别研究

Cross⁃temporal Mental Workload Identification Based on Transfer Learning Boosting

在航天任务中对脑力负荷的有效检测可以保障任务执行效率和生产安全,预测航天员的表现。针对跨时间脑力负荷检测,依据MATB任务设计了包含不同难度任务的跨时间实验范式,采集了14名志愿者3次不同时间的脑电数据。基于迁移学习Boosting方法,引入辅助数据,设计了基于TrAdaboost的跨时间脑力负荷识别算法,在没有目标数据参与的情况下进行了跨时间的分类识别。探索了最佳分段长度和辅助样本比例对识别效果的影响,并基于多个数据样本进行了决策研究,跨时间下的脑力负荷最佳识别准确率达到74.73%。结果表明,提出的跨时间脑力负荷分类框架实现了脑力负荷的有效识别。

The effective detection of mental workload in space missions can ensure the efficiency of mission execution and work safety,and predict the performance of astronauts.To detect the cross-temporal mental workload,an experimental paradigm that includes tasks of varying difficulties was designed based on the MATB task and EEG data was collected from 14 subjects at three different time points.Based on the transfer learning Boosting method,a cross-temporal mental workload rec-ognition algorithm was designed based on TrAdaboost by introducing auxiliary data and cross-time classification and recognition were carried out without the participation of target data.In addition,the influence of the optimal segment length and the proportion of auxiliary samples on the recognition effect were explored and decision-making research was conducted based on multiple data samples.The optimal recognition accuracy for mental workload across time reached 74.73%.The results demonstrate that the cross-temporal mental workload classification framework proposed in this study can effectively identify the mental workload.