The wavelengths
of the extracted spectral data had a high degree of
dimensionality with redundancy among continuous wavelengths,
resulting in the slow speed of hyperspectral line-scanning.
Although multivariate analysis can be used to process such abundant
data, it is usually time-consuming and can lead to over-fitting.
Therefore, it was effective to reduce the large variables into several
optimal ones carrying the most valuable and effective information,
thus simplifying the calibration process and enhancing accuracy
and robustness of the established models (Liu, Sun, & Zeng, 2014).