This thesis study employs textural classification techniques applied to four different data groups: (1) visible light photography, (2) high-resolution drill core line scan imaging (3) scanning electron microscopy backscattered electron (SEM-BSE) images, and (4) 3D data from X-ray microtomography (μXCT). Eleven textural classes from Aitik ores were identified and characterized. The distinguishing characteristics of each class were determined such as modal mineralogy, sulphide occurrence and Bond work indices (BWI). The textural classes served as a basis for machine learning classification using Random Forest classifier and different feature extraction schemes. Trainable Weka Segmentation was utilized to produce mineral maps for the different image datasets. Quantified textural information for each mineral phase such as modal mineralogy, mineral association index and grain size was extracted from each mineral map. Efficient line local binary patterns provide the best discriminating features for textural classification of mineral texture images in terms of classification accuracy. Gray Level Co-occurrence Matrix (GLCM) statistics from discrete approximation of Meyer wavelets decomposition with basic image statistical features[PK1] (e.g. mean, standard deviation, entropy and histogram derived values) give the best classification result in terms of accuracy and feature extraction time. Differences in the extracted modal mineralogy were observed between the drill core photographs and SEM images which can be attributed to different sample size[PK2] . Comparison of SEM images and 2D μXCT image slice shows minimal difference giving confidence to the segmentation process. However, chalcopyrite is highly underestimated in 2D μXCT image slice, with the volume percentage amounting to only half of the calculated value for the whole 3D sample. This is accounted as stereological error. Textural classification and mineral map production from basic drill core photographs has a huge potential to be used as an inexpensive ore characterization tool. However, it should be noted that this technique requires experienced operators to generate an accurate training data especially for mineral identification and thus, detailed mineralogical studies beforehand is required. / Primary Resource Efficiency by Enhanced Prediction (PREP) / Center for Advanced Mining and Metallurgy (CAMM)
| Identifer | oai:union.ndltd.org:UPSALLA1/oai:DiVA.org:ltu-65207 |
| Date | January 2017 |
| Creators | Tiu, Glacialle |
| Publisher | Luleå tekniska universitet, Mineralteknik och metallurgi, EMerald Program |
| Source Sets | DiVA Archive at Upsalla University |
| Language | English |
| Detected Language | English |
| Type | Student thesis, info:eu-repo/semantics/bachelorThesis, text |
| Format | application/pdf |
| Rights | info:eu-repo/semantics/openAccess |
Page generated in 0.0059 seconds