Continuous Wavelet Transformations for Hyperspectral Feature Detection

A novel method for the analysis of spectra and detection of absorption features in hyperspectral signatures is proposed, based on the ability of wavelet transformations to enhance absorption features. Field spectra of wheat grown on different levels of available nitrogen were collected, and compared to the foliar nitrogen content. The spectra were assessed both as absolute reflectances and recalculated into derivative spectra, and their respective wavelet transformed signals. Wavelet transformed signals, transformed using the Daubechies 5 motherwavelet at scaling level 32, performed consistently better than reflectance or derivative spectra when tested in a bootstrapped phased regression against nitrogen.