A Novel Exploratory Shape Analysis Method to Identify Spatial Variation of Fungi Infection Level on Wheat Kernels

摘要

In hyperspectral imaging the raw data is a measurement of one sample at different energy bands to form multiple images, typically hundreds. If images from different energy bands contain the same spatial pattern (i. e., shape) then the biochemicals present in the sample are typically identifiable from these energies. A novel method to determine if two images contain the same shapes and show the matching areas is introduced here. This is particularly of interest in applications where no reference visual image is available. For example, in data collected using a visually opaque Attenuated Total Reflection (ATR) crystal, it is impossible to determine where the crystal contacts the sample. Fusarium infected Canadian red spring wheat kernels with appropriate controls were imaged at different stages of infection using a Fourier transform mid-infrared microscope equipped with a germanium ATR accessory. The extreme conditions were examined and all spectral peaks were integrated to establish a set of integration parameters for the entire data set. The preliminary results indicate that the measure of shape similarity developed here is well approximated by the correlation of the original integrated grayscale images. This allows one to quickly compute which images to compare and explore the mean spectra of the matching areas in order to determine those regions with the highest infection levels and the presence of toxic secondary metabolites.