|
 |
|
 |
UQ Innovation Expo 2003 |
|
ITEE Innovation Expo 2002 |
|
ITEE Innovation Expo 2001 |
|
ITEE Public Web |
|
ITEE School Alumni |
| UQ Innovation Expo 2003 » Exhibits - by Industry Sector » Paul Whiten |
Machine Learning for Mineral Classification
Exhibitor: Paul Whiten
Supervisor: Marcus Gallagher
Research Group: Complex and Intelligent Systems
Industry Sector: Manufacturing and Industrial Applications
A neural network is an artificial intelligence (AI) programming technique that displays learning capabilities. This ability makes them ideal for applications involving the prediction and classification of data. One potential use for neural networks is in the identification of minerals from ore samples. The goal of this thesis is to expand on previous students' work, in an effort to create an automated mineral identification system. The system would apply a neural network to determine how many mineral species exist in a sample and classify them according to X-ray spectrum data. The implications of such a system would be the potential use in industry as an alternative classification method to traditional statistical techniques. A successful solution would have to be fast, reliable and accurate.
The model for a neural network is based upon the operation of the human brain. Cells, called neurons, receive data, process it and output it to further neurons. This 'network of neurons' is capable of being trained to classify unfamiliar data and make informed decisions. Previous work focused on using neural networks to classify a single mineral type. However an ore sample can contain multiple (sometime up to 200) mineral species. This thesis focuses on improving the single mineral identification methods and expanding the developed neural network to accommodate the classification of multiple mineral types.
The process starts by taking an unknown mineral sample and analysing it using an electron microscope to produce an X-ray spectrum. Each mineral has a unique X-ray signature that can be used for identification. The spectrum data is then processed through normalisation and feature extraction, which is used to determine the most influential characteristic factors in the data (features could be, for example in the case of humans – height, weight, age etc). The pre-processed data is then input into a neural network that classifies the mineral type.
