Comparison of spectral matching algorithms for identifying natural salt crusts

This paper presents a comparison between the capabilities of different spectral matching algorithms to indentify the spectra of samples from salt crusts obtained by a hand-operated spectroradiometer. The following algorithms have been used: absolute difference; squared difference; derivative difference; Euclidean vector distance; and correlation coefficient. As the investigations have shown, the results depend on the method used. For example, the correlation coefficient algorithm yielded the best results, next come the squared derivative, the Euclidean vector difference, and the absolute difference algorithms. It has been found that automated search of spectra of salt crusts does not permit exact identification of the unknown spectra of salts by means of the existing spectra libraries. Besides the quality of the algorithms and libraries used, the discrepancy between the results of matching is also due to the inherent factors related to the nature of the salt crusts, such as variations in the layering, moisture content, grain size, etc. However, in the case of large databases of the spectra of salts or minerals, the above algorithms can be useful for reducing the search to a small number of spectra of interest which should subsequently be interpreted in a conventional manner.