Classification of six sound categories using functional near-infrared spectroscopy

Abstract

Sound category recognition is essential for auditory perception in the temporal cortex. This paper aims to classify six different sound categories based on functional near-infrared spectroscopy (fNIRS). Recursive least square estimation was applied to remove physiological noises, and various features were extracted from the oxy-hemoglobin (HbO) activation curve in the time domain. Extracted features were selected by statistical method to maximize the differences between sound categories. Furthermore, various classification methods (linear discriminant analysis (LDA), support vector machine (SVM), k-nearest neighbor (kNN), and naïve Bayes classifier (NB)) were applied to get the acceptable accuracy. The classification accuracies were 20.17% with LDA, 32.61% with SVM, 24.54% with kNN, and 25.71% with NB. This work demonstrates the potentiality of the online classification process for decoding what people hear using fNIRS.