Sparse coding classification and cepstral singular value for cognitive workload estimation. (May 2021)

• Record Type: Journal Article
• Authors: Ghaderyan, Peyvand; Abbasi, Ataollah
• Is Part Of: Computers & electrical engineering. Volume 91(2021)
• Publication Date: 2021-05
• Subjects: Electrocardiogram -- Arithmetic task -- Workload estimation -- Pattern recognition; Computer engineering -- Periodicals; Electrical engineering -- Periodicals; Electrical engineering -- Data processing -- Periodicals; Ordinateurs -- Conception et construction -- Périodiques; Électrotechnique -- Périodiques; Électrotechnique -- Informatique -- Périodiques; Computer engineering; Electrical engineering; Electrical engineering -- Data processing; Periodicals; Electronic journals; 621.302854
• DOI: 10.1016/j.compeleceng.2021.107031
• Languages: English
• ISSNs: 0045-7906

Abstract

Highlights: A new ECG measure for automatic cognitive load classification is proposed. The measure integrates the quefrency and algebraic information of the signals. A sparse coding method is successfully developed to classify workload levels. The proposed method outperforms previous ECG-based workload estimation algorithms. Abstract: Cognitive workload estimation (CLE) is an interesting but challenging task with applications ranging from diagnosis and treatment of nervous system disorders to brain computer interface. The performance of CLE is usually affected by individual differences and recording noises. Noise-robustness paradigms that are independent of small variations may offer a solution to this challenge. Hence, a new CLE system based on the novel application of integrated singular value decomposition (SVD), cepstrum analysis and sparse non-negative least-square coding method was proposed. It extracted both of algebraic and harmonic information. The method was tested using electrocardiogram signals of 45 subjects while performing an arithmetic task with different levels of difficulty. It effectively estimated the workload levels using a small number of features with an average accuracy of 91%. The Hankel matrix-based SVD performed as well as non-overlapping matrix. Furthermore, significant improvement in the performance was observed as compared to conventional classifiers and features.