What question did this study set out to answer?

This research aims to evaluate the effectiveness of different unmixing models for estimating mineral abundances from hyperspectral data.

March 26, 2020

Estimation of Mineral Abundance From Hyperspectral Data Using a New Supervised Neighbor-Band Ratio Unmixing Approach

Key Points

This research aims to evaluate the effectiveness of different unmixing models for estimating mineral abundances from hyperspectral data.
Compared nine unmixing models including a new approach, neighbor-band ratio unmixing (NBRU), on 94 mineral mixtures.
Used radiative transfer models as benchmarks in performance comparisons.
Assessed performance on an AVIRIS hyperspectral image of the Cuprite area, NV, USA.
NBRU demonstrated the lowest mean and median errors at 9.8% and 7.4%, respectively.
Radiative transfer models showed better estimates, with errors of 6.5% and 5.6%, and 6.7% and 4.7% for Hapke's and Shkuratov's models, respectively.
NBRU provided the best spatial distribution mapping for seven minerals compared to validation maps.

Abstract

This article compares the ability of nine unmixing models, including radiative transfer (RT) models as well as a new nonlinear unmixing approach called neighbor-band ratio unmixing (NBRU), to obtain mineralogical information from hyperspectral data. Their performance in estimating mineral abundances of 94 crafted mineral mixtures was first assessed. NBRU led to the best results among non-RT models with mean and median errors of 9.8% and 7.4%, respectively. Hapke's and Shkuratov's RT models obtained 6.5% and 5.6%, and 6.7% and 4.7%, respectively. In a second experiment, the mapping ability of six non-RT models and their robustness when facing endmember variability were evaluated. The assessment was performed on an AVIRIS hyperspectral image of the widely studied Cuprite area, NV, USA. Comparisons with validation maps showed that NBRU retrieved the best spatial distributions for seven of the nine minerals mapped.

Demander à l'IA

Bookmark