Abstract:

Based on SVM statistical theory, which was improved from kernel function’s construction，the original Hyperion hyperspectral data were dimensionally reduced, transformed and feature extracted by means of principle component analysis, envelope removal and spectra derivative transformation. The different regression results through the transformations were analyzed and compared. Then it was applied in the retrieval of rock’s oxide weight percent in Huolinguole, Inner Mongolia. A new mineral quantitative retrieval method was proposed for some important mineral which has not obvious characteristics of the spectral curve. Based on the hyperspectral remote sensing data, the chemical composition of mineral was inverted by improved SVM regression technology. Through the CIPW, the standard mineral percentage composition was derived. The results of the study show that：the SVM regression accuracy improved by improving the nuclear function, and the derivative of the transformed inversion precision up to 74.87%, improved 4.11% comparing to the original spectrum inversion precision. CIPW performed well in geological mapping using hyperspectral remote sensing, and provides a scientific basis on identification and evaluation of lithology.

Key words: support vector machines, rock, norm mineral calculation, minerals