Artificial neural network (ANN) modelsare a common tool forcognitive state assessment. It is best if the inputs to the model are periodic. Typically, these inputs are extractedfrom physiological signals such as the electroencephalogram (EEG), electrocardiogram (ECG), electrooculogram (EOG), and others. Spectral measures derived from EEG data are periodic due to the signal processing. Features based on heart activity and respiration are quasi-periodic by nature. Features extracted from EOG, such as blink rate, can be especiallynon-periodic and can contain outliers. One approach to deal with this problem is to use static windows to compute average blink rate. This approach has some shortcomings. A new approach that uses dynamic windowing, filtering, and sampling is presentedhere. Thisnew approach produces periodic data that aredynamic, adaptive to the individual, and well suited for ANN model use.
& Galster, S.
(2017). Periodic Blink Measures Using Dynamic Windowing. 19th International Symposium on Aviation Psychology, 524-529.