摘要：A model is presented that performs spectral deicing of mixed pixels in satellite images of glaciated areas. The model was tested in Northern Victoria Land, Antarctica. For this region we assumed that pixels could be grouped in two broad categories representing pure ice and pure rock. Naturally mixed ice and rock pixels are present in satellite images; these were recomputed to separate the spectral component related to the rock fraction. We used Landsat TM images as input data and aerial photographs, maps and field surveys as reference data. By making use of sample populations of pixels corresponding to pure ice and to pure rock groundels (i.e. ground elements, the ground portions corresponding to each pixel), we detected the linear correlation between pairs of bands and selected the two most suitable bands. For every pixel falling between the correlation lines of the two categories, the rock fraction in the corresponding groundel was computed. This fractional value was then used to perform the automatic deicing process with which the DNs of the selected mixed pixels were recomputed. In the utilised Antarctic image, this process increases about 2.7 times the number of pixels in the pure rock category, allowing the production of enhanced images and, as a side product, a thematic map of rock percentage in the groundels.
