Transforming the Mining Value Chain
T2a CORESCAN DATA FEATURE EXTRACTION AND CLASSIFICATION FOR MINERALOGICAL AND TEXTURAL INFORMATION ANALYSIS
LEADER: | |||
Matthew Cracknell | |||
TEAM MEMBERS: | |||
Ron Berry, Neil Goodey, Anthony Harris, Anya Reading | |||
STUDENTS: | |||
Javier Merrill Angela Rodrigues | CODES Monash University | ||
COLLABORATORS: | |||
Ekaterina Savinova Laurent Ailleres, Robin Armit | Corescan Monash University |
PROJECT SUMMARY
2019
Corescan generates a range of drill core image products including Digital Surface Models (DSM), Red-Green-Blue (RGB) colour photographs and Visible-Near Infrared–Short Wave-Infrared (VNIR–SWIR)-derived mineral interpretations. Despite the rich geological information implicitly contained within these data, they are primarily used to provide percentages of identified minerals down hole to Corescan customers. The aim of this project is to classify and extract mineralogical and textural features from Corescan imagery, adding value to their data products. For example, the two-dimensional imagery generated by Corescan contains information on the geometric characteristics and spatial arrangement of interpreted minerals, while there are key economic mineral species, such as sulfides, that do not have characteristic absorption features in VNIR–SWIR spectra and are therefore not identified accurately.
In 2019 a significant boost was the addition of PhD candidates Javier Merrill (CODES) and Angela Rodrigues (Monash) to this project. The PhD projects are embedded within AMIRA P1202 Module 4 and are investigating novel approaches to obtaining numeric representations of mineral texture for geometallurgical domaining and the use of machine learning for improving hyperspectral mineral classifications. Dr Matthew Cracknell presented at PACRIM 2019 in Auckland, New Zealand, highlighting research on identification of sulfides and quantification of mineral textures in Corescan RGB photographs. In April and November, Matthew co-hosted one-day introduction to machine learning workshops as part of the AusIMM PACRIM 2019 (Auckland, New Zealand) and Mining Geology 2019 (Perth, Australia) conferences. The aim of these workshops was to introduce the fundamental concepts of machine learning and encourage uptake in the geoscience community. Matthew has also run a number of data analytics workshops for CODES/TMVC researchers.
2018
Corescan generates a range of drill core image products including Digital Surface Models (DSM), Red-Green- Blue (RGB) colour photographs and Visible-Near Infrared–Short Wave-Infrared (VNIR–SWIR)-derived mineral interpretations. Despite the rich geological information implicitly contained within these data, they are primarily used to provide percentages of identified minerals down hole to Corescan customers. The aim of this project is to classify and extract mineralogical and textural features from Corescan imagery that add value to their data products. For example, the two-dimensional imagery generated by Corescan contains information on the geometric characteristics and spatial arrangement of interpreted minerals, while there are key economic mineral species, such as sulfdes, that do not have characteristic absorption features in VNIR–SWIR spectra and are therefore not identified accurately in available data products.
In 2018, project outputs included an article published in the open access journal, Minerals, titled ‘Automated Acid Rock Drainage Indexing from Drill Core Imagery’ that combined research on identification of sulfides and quantification of mineral textures in Corescan RGB photographs. This research solidifies collaborations between Corescan staff and Theme 3 TMVC researchers and PhD students, and documents a practical application of feature extraction and analysis for rapidly and consistently estimating indices of rock acid forming potential. Matthew Cracknell was invited to speak at the ‘AI/ Machine Learning; Opportunities and Challenges for Mineral Exploration’ workshop at AEGC2018 in Sydney. His presentation, ‘Unsupervised Clustering of Geoscience Data’, guides those new to the application of machine learning methods to mineral exploration on best-practice data handling, processing and analysis. With close to 50 participants, this workshop was one of the more popular post-conference events at AEGC2018. In October, Matthew Cracknell and several PhD students hosted a three-day workshop for First Quantum Minerals (FQM). The aim of this workshop was to introduce the fundamental concepts of machine learning and encouraging uptake of TMVC developed algorithms and workflows for mineral exploration by FQM employees.
2017
Corescan™ generates a range of drill core image products including Digital Surface Models (DSM), Red-Green-Blue (RGB) colour photographs and Visible-Near Infrared–Short Wave-Infrared (VNIR–SWIR)-derived mineral interpretations. Despite the rich geological information implicitly contained within these data, they are primarily used to provide percentages of identified minerals down hole to Corescan™ customers. The aim of this project is to classify and extract mineralogical and textural features from Corescan™ imagery that add value to their data products. For example, the 2-dimensional imagery generated by Corescan™ contains information on the geometric characteristics and spatial arrangement of interpreted minerals, while there are key economic mineral species, such as sulfdes, that do not have characteristic absorption features in VNIR–SWIR spectra and are therefore not identified accurately in available data products.
In 2017, outputs included a presentation at the International Association for Mathematical Geosciences (IAMG) 2017 conference titled ‘Image texture quantification from Corescan™ mineral classifications’ which outlined research into quantifying mineral textures. This research forms the basis for developing meaningful indices characterising geological textures, such as disseminated versus massive and degree of veining. Dr Cracknell was also invited to chair two sessions on ‘Machine Learning, Pattern Recognition, Data Mining, Big Data’ at IAMG2017. These presentations were well received by both the public and expert geoscientists alike.
In September, recent progress on research into robust classification of sulfides from Corescan™ imagery and subsequent analysis of sulfide textural characteristics was presented to Corescan™ staff. Applications of this work for the automated classification of Acid Rock Drainage risk are being developed in collaboration with Corescan™ staff and Theme 3 researchers. Relevant algorithms are in an advanced testing phase and relevant manuscripts will be ready for submission in early 2018.
2016
This project aims to extract and classify mineralogical and textural features from current Corescan™ output datasets. The project will primarily use Corescan™ visible and hyperspectral imagery, and surface geometry imagery, as input data to extract and classify features, such as veins, fractures, grain boundaries, lithologies, alteration, textural fabrics and ARD domains. Output features will add value to the dense datasets currently generated by the Corescan™ system by making them more geologically interpretable, and thus accessible to users. Features will also be used to support quantitative analyses and qualitative interpretations across the TMVC’s three themes – covering footprints, geometallurgy and geoenvironment.
During the year, a prototype fracture detection algorithm was developed to identify and extract fractures from Corescan™ data. The outputs of this algorithm feed directly into workflows being developed for the extraction of geotechnical parameters.