The scanning electron microscope has been used for determining mineral composition and gathering geometric data of particles for mineral liberation analysis for decades. With the increasing power of computers, it is possible to further extend the use of scanning electron microscopy to achieve more accurate and faster mineral liberation analysis in real time. In such an effort, it is necessary to develop algorithms to improve back-scattered electron image segmentation to generate more accurate mineral phase maps. X-ray spectrum analysis using least square fit, which is widely used in applications in many fields, is relatively mature and offers better accuracy in terms of X-ray energy and intensity. Adapting the X-ray spectrum analysis techniques for a scanning electron microscope and linking the X-ray spectrum analysis to mineral identification become another necessary part.
Algorithms are developed for image segmentation and
mineral identification respectively. The proposed image segmentation algorithm measures the "significance" of the grey-level uniformity of a region and evaluates the "effective grey-level difference" between two adjacent regions with a scale control built in. Then it uses a multi-level iteration of region-merging to overcome the multi-scale problem. A pattern matching algorithm is proposed to match the line series obtained by least square fit from the X-ray spectrum of unknown mineral, to the standard line series of minerals stored in the mineral database. It traverses all possible matches from the sample line series to standard line series in the database for the best match. A null line is introduced to cope with the problem that occurs when line series contain false lines and two series have different number of lines.
The experiments of both algorithms give satisfactory results in most cases that are considered
"normal" in actual measurement. At very high magnification, the limitations of the segmentation algorithm become obvious, such as crack and shadow extending unexpectedly and scalability getting worse. The pattern matching algorithm for X-ray line series match fails to distinguish haemetite and magnetite, which have close elemental composition. The difference in the iron content is about 2%. It also fails to distinguish diopside and Cr-diopside. The chromium content in Cr-diopside is 0.34%, which is too low for the characteristic X-ray of Cr to be detectable in the spectrum. The chromium content does not have any detectable impact on the total elemental composition of Cr-diopside as well. With 52 minerals tested, except for the above two mineral pairs, all other minerals tested are reliably distinguished.
Chapter one gives an introduction on mineral identification using an SEM and literature reviews on both image segmentation and X-ray
spectrum analysis applications. The problems in developing a new segmentation algorithm and applying X-ray spectrum analysis to mineral identification using a SEM are also addressed in chapter one. Chapter two discusses the algorithm development for BSE image segmentation. The experiment of the proposed segmentation algorithm is discussed in chapter three. Chapter four discusses the methodologies for applying X-ray spectrum analysis to SEM. Problems specific to mineral identification using an SEM, are also addressed. A pattern match algorithm is proposed to match standard line series and sample line series. Chapter five discusses the experiment of the pattern match algorithm. Chapter six summarises the experiment results and discusses conclusions.