Application of Machine Learning Techniques to Mineral Recognition

Student: Peter William Deacon

Supervisor: Dr. Marcus Gallagher

Category: Computer Systems Engineering Thesis Project

The task of classifying minerals from x-ray spectra is a challenging problem using conventional statistical techniques. There are several reasons why they are not completely successful including noise and when x-ray spectra are mixtures of two or more mineral spectra. Machine learning techniques provide an alternative solution to this problem and can also accurately classify mineral spectra. The aim of the research is to improve a mineral analysis system developed by JKTech, which is a commercial division of UQ’s JKMRC.

Techniques investigated and used during analysis include the multi-layer perception, naive bayes, regularised adaboost, principal component analysis and Kohonen self-organising maps. Results show that several machine learning techniques can classify x-ray spectra but certain techniques are more successful than others. These techniques provide an excellent alternative for classifying x-ray spectra, especially as a fallback when conventional pattern matching techniques are unsuccessful in classifying the spectra.

Poster Presentation (PDF)

Thesis Document (PDF)