Extraction of mineral absorption features from high-spectralresolution data using non-parametric geostatistical techniques

High spectral resolution images (AVIRIS) providing detailed information on the surfaced mineralogy have been used to evaluate a new indicator kriging based absorption feature extraction technique for mineral mapping exploiting the spectral characteristics of image pixel spectra in comparison with laboratory spectra of minerals. Raw radiance image data from the 1989 Cuprite Mining data set were corrected for atmospheric influence and solar irradiance drop off reducing the data to reflectance using the Empirical Line method. A new technique for mineral mapping is presented based on ordinary indicator kriging. The approach directly uses mineral spectra by defining key bands on shoulders and centres of absorption features which are automatically detected from image spectra using statistical zonation techniques. The image data of selected spectral bands are transformed into binary (0,1) images: the pixel value becoming I if the pixel DN value falls within upper and lower limits characterizing the mineral absorption feature in the spectral band. Performing ordinary kriging and integrating the estimates for the selected bands yields the probability that a mineral absorption feature is present or absent for each pixel (or subpixeJ area) in the scene. Furthermore, kriging provides an estimate for the reliability of this probability.