Successes and challenges of factor analysis/target transformation application to visible-to-near-infrared hyperspectral data

Abstract We designed a laboratory visible-to-near-infrared (VNIR) hyperspectral experiment to test the effectiveness of factor analysis/target transformation for detecting minerals mixed with Mars Global Simulant-1 (MGS-1). The purpose of this experiment is to test for true positive, true negative, false positive, and false negative results from application of factor analysis/target transformation methods and determine the parameters that dictate good versus bad algorithm performance. Gypsum, calcite, montmorillonite, nontronite, and kaolinite were each mixed with MGS-1 at abundances of 1%, 2.5%, 5%, 10%, 20%, and 50%. The mixtures were placed in 2.5 × 2.5 × 1 cm sample trays and imaged using a Headwall Imaging Spectrometer with a spectral range of 0.9–2.6 μm, 8.98 nm spectral sampling, and 0.34 mm/pixel spatial resolution. These images include thousands to tens of thousands of hyperspectral pixels covering each individual mixture tray. Full-image factor analysis/target transformation (FA/TT) and Dynamic Aperture Factor Analysis/Target Transformation (DAFA/TT) were applied to these data to detect the minerals mixed with MGS-1. The results demonstrate that factor analysis/target transformation is prone to both false positive and false negative detections, but in certain applications—including DAFA/TT—it can be useful for highlighting spectrally interesting areas in hyperspectral images for follow-up investigation. The results presented here demonstrate that applications of factor analysis/target transformation to VNIR hyperspectral datasets should be used to highlight small outcrops and/or weak spectral signals in pixels for follow-up investigation. This emphasizes the need for supporting evidence to be obtained—in addition to factor analysis/target transformation—before interpretations of planetary surface processes should be made.