A new signal analysis method for functional near-infrared spectroscopy

For research and application of functional near-infrared spectroscopy in neuroscience, appropriate signal analysis method of functional near-infrared spectroscopy is significant. Most of researchers have applied traditional statistical features for feature extraction, and classic machine learning method like support vector machine for further analysis. In this paper, a new feature extraction method based on principles of multivariate graphic representation has been proposed. Then, supervised sparse representation based on partial order structure theory has been suggested for signal pattern classification. Both methods have been tested by signal analysis experiment of functional near-infrared spectroscopy, which is designed to achieve mental workload assessment during n-back work memory experiment. The result indicated, new methods of this work could be applied in functional near-infrared spectroscopy feature extraction and signal analysis.