Integrated Analysis of Carbonatite using Short Wave Infra-Red and Visible/Near Infra-Red Reflectance Characteristics
Abstract
Carbonatites are among the most important hosts for economically important rare-earth element (REE) deposits. An ongoing challenge has been the identification of carbonatites, which may outcrop as small bodies with indistinct field characteristics. Remote sensing techniques may provide a routine and reliable method to identify such deposits. We have used short wave infra-red (SWIR) and visible/near infra-red (VNIR) reflectance characteristics of a well exposed carbonatite located in the north east of the United Arab Emirates to develop techniques to facilitate the distinction of carbonatites from other rock types. This project has focused on the wavelength region from 0.45 to 2.43 μm of SWIR and VNIR subsystems on the imaging instrument onboard ASTER. We hypothesize that based on spectral and spatial data derived from computer segmentation algorithms of the SWIR (7) and VNIR (4, 1) bands we will be able to identify carbonatite rocks. In order to build a technique that can capture the intrinsic associations between carbonatite anomalies, rock class types, and attributes, we assembled the spectral and spatial information derived from computer segmentation algorithms into a single segment image. During our investigation of the spatial data two principal questions arose: 1) How should spatial structures, or "neighborhoods" for each pixel within the image be automatically defined? 2) How should spatial and spectral information be combined in the classification? We addressed these questions by using unsupervised and supervised segmentation algorithm strategies based on pixel values and locations. Pixels that are spatially connected and have similar values were grouped in a single segment (fixed neighborhood pixels) on the basis of the integration of the maximum-likelihood supervised classification technique within a Markov Random Fields framework. We then developed guidelines for combining the spatial information extracted through segmentation with spectral information within a classifier. These techniques allowed us to devise classification techniques using region-growing segmentation based upon automatically derived markers for identifying spatial structures of carbonatite. We were able to develop methods that are able to handle high-dimensional data and use adaptive spatial neighborhoods derived from segmentation results. The final strategy concentrated on techniques to reduce over-segmentation of the resulting images. An important component of this strategy consisted of analyzing potential classification results for the most reliably classified pixels to be used as markers of spatial zones. We present our results as a thematic image in which pixel values represent a specific spectra signature of class types of contiguous raster objects of carbonatites materials.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2012
- Bibcode:
- 2012AGUFMIN23A1495A
- Keywords:
-
- 1926 INFORMATICS / Geospatial;
- 1980 INFORMATICS / Spatial analysis and representation