On the development of a hyperspectral library for autonomous mining systems

The use of hyperspectral imagery for robotics is predicted to increase as costs of sensors decline. A library of spectra can be used to map hyperspectral data to identify objects by comparing their reflectance signature to known materials. In this paper, methods used to build the spectral library to map geology on mine faces are described. The library includes various (simulated) environmental conditions such as different light sources and the inclusion of shade and moisture. The principal focus of this paper is the inclusion of shade and moisture into the spectral library and to investigate their effects on curve shape and albedo. These effects are usually not considered in other spectral libraries. The signal-to-noise ratios (SNRs) are greatest for spectra acquired under artificial light and become progressively smaller for spectra acquired under natural light, moisture and shade. Shade decreases the brightness but does not generally alter the shape of the spectral curve. Moisture decreases albedo but less than shade does, however, moisture changes the shape of the spectral curve. Principal component analysis suggests that several major rock types could be distinguished on the basis of their spectral reflectance. This study demonstrates the importance of collecting library spectra under a range of conditions in order to achieve an accurate mapping of covertypes.