Mineral Identification And Characterization: An Integrated Approach To Recover Mineralogical Information From Hyperspectral Images
Résumé
Mineral mapping and soil analysis, from airborne or satellite sensors, are two growing topics for the study of environmental issues and the evaluation of the geological interest of large areas. For example, it can provide important information for the supervision of industrial activities. The development of new and more performing imaging spectrometers such as AVIRIS-NG or EnMAP, makes possible the acquisition of hyperspectral images in the Visible Near-InfraRed [0,4 – 1,3] μm and the Short-Wave InfraRed [1,3 – 2,5] μm domains with high spectral resolution and signal-to-noise ratios allowing the analysis of the reflectance spectra of minerals. Indeed, those spectral signatures may give access to several physical and chemical parameters (e.g., composition, grain size, humidity) of minerals.
The purpose of this poster is to present an integrated approach giving the possibility to a non expert to automatically study the mineralogy of an area from a hyperspectral image. It is based on the AGM (Automatized Gaussian Model) procedure [1] for which several evolutions have been developed. This procedure uses the EGO (Exponential Gaussian Optimization) model which decomposes the reflectance spectra of minerals as the sum of a continuum and generalized Gaussians functions. Each Gaussian, defined by five parameters (center, amplitude, width, asymmetry and saturation), represents an absorption band of the spectrum. Using this model, several physical and chemical parameters of minerals can be estimated.
The new AGM procedure is divided in four main parts. A first step, based on the estimation of the signal dependent and independent noises in the imaging spectrometer, gives information about the measurement uncertainties. Then, the parameters of the EGO model are estimated. A nonlinear least-square solver, such as Levenberg-Marquardt or Optimal Estimation, is considered depending on the a priori knowledge (geology of the area, measurement uncertainties, etc.). A third step, using a fuzzy logic system developed at CEA [2], allows, from the estimated model parameters and considering a set of expert rules, to identify the diagnostic absorption features of the reflectance spectra and so the mineral or the mixture of minerals associated to it. Thus, the database consists in absorption features rather than reflectance spectra. Finally, several physical and chemical parameters of the minerals can be mapped (e.g., composition, grain size) from the mineral identification and some of the model parameters.
This new approach, automatic and adapted to non-experts operational needs, is validated on AVIRIS and AVIRIS-NG data for areas of geological interest. It allows to identify mixtures of minerals and gives access to several of their physical and chemical parameters. Future work will include the development of new optimization methods and the creation of new expert rules.
[1] R. Marion and V. Carrère, Mineral Mapping Using the Automatized Gaussian Model (AGM) - Application to Two Industrial French Sites at Gardanne and Thann, Remote Sensing 2018, 10, 146.
[2] J.P. Poli and L. Boudet, A fuzzy expert system architecture for data and event stream processing, Fuzzy Sets and Systems, 2018, 343, p.20-34
Origine : Fichiers produits par l'(les) auteur(s)
Loading...