An ELM-based Deep SDAE Ensemble for Inter-Subject Cognitive Workload Estimation with Physiological Signals

Evaluating operator cognitive workload (CW) levels in human-machine systems based on neurophysiological signals is becoming the basis to prevent serious accidents due to abnormal state of human operators. This study proposes an inter-subject CW classifier, extreme learning machine (ELM)-based deep s...

Celý popis

Uložené v:
Podrobná bibliografia
Vydané v:Chinese Control Conference s. 6237 - 6242
Hlavní autori: Zheng, Zhanpeng, Yin, Zhong, Zhang, Jianhua
Médium: Konferenčný príspevok..
Jazyk:English
Vydavateľské údaje: Technical Committee on Control Theory, Chinese Association of Automation 01.07.2020
Predmet:
Cognitive workload
EEG
Electroencephalography
ensemble learning
extreme learning machine
Fuses
Noise reduction
Physiology
stacked denoising autoencoder
Time measurement
Training
ISSN:1934-1768
On-line prístup:Získať plný text
Tagy: Pridať tag
Žiadne tagy, Buďte prvý, kto otaguje tento záznam!
Popis
Shrnutí:Evaluating operator cognitive workload (CW) levels in human-machine systems based on neurophysiological signals is becoming the basis to prevent serious accidents due to abnormal state of human operators. This study proposes an inter-subject CW classifier, extreme learning machine (ELM)-based deep stacked denoising autoencoder ensemble (ED-SDAE), to adapt the variations of the electroencephalogram (EEG) feature distributions across different subjects. The ED-SDAE consists of two cascade-connected modules, which are termed as high level personalized feature abstractions and abstraction fusion. The combination of SDAE and locality preserving projection (LPP) technique is regarded as base learner to obtain ensemble members for training meta-classifier by stacking-based approach. The ELM model with Q-statistics diversity measurement is acted as meta-classifier to fuse above inputs to improve classification performance. The feasibility of the SD-SDAE is tested by two EEG databases. The multi-class classification rate achieves 0.6353 and 0.6747 for T1 and T2 respectively, and significantly outperforms several shallow and deep CW estimators. By computing the main time complexity, the computational workload of the ED-SDAE is also acceptable for high-dimensional EEG features.
ISSN:1934-1768
DOI:10.23919/CCC50068.2020.9188806