Boundary determination of mineral regions in hyperspectral drill core imager data

Abstract

This work is about segmentation of hyperspectral images in order to identify the different regions within the image and determine their best boundaries. A new method based on the Fitzhugh-Nagumo model is introduced to do this by extracting additional useful information from the image data. It is inspired by the work done in [1] titled “A novel approach to text binarization via a diffusion-based model”, which was applied on 2-dimensional images in grayscale. The results from the proposed method is then assessed and compared to other existing methods on a scene of Cuprite, Nevada, as well as drill core imager data. The datasets chosen for this research are the artificial and real AVIRIS Cuprite of Nevada and as well as real drill core imager data. Cuprite, Nevada was chosen because it is well studied and other methods have been applied to it. The core imager data is studied due to its importance of delineating minerals regions and the likelihood of getting purer pixels given the high spatial resolution of the data.