Neural networks and the classification of mineralogical samples using x-ray spectra

The automatic classification of large numbers of mineral samples is a practical problem in mining research. A system currently in use is based on simple statistical tests. Although the system performs well under typical conditions, the data collection procedure can be very time-consuming. This time can be significantly reduced, but at a cost of introducing noise into the data, leading to a degradation in classification performance. This paper reports on an initial investigation into the application of neural network techniques to the mineral identification task, and compares the performance of these methods to the current system. The results are very encouraging and suggest that a more powerful classifier might allow die data collection process to be significantly sped up without significant loss of classification accuracy for the overall system.