Sandra Lorenz WiMLDS Dresden
0 likes123 views
This document discusses how data science and machine learning can help achieve more sustainable mineral exploration. It outlines several challenges with conventional exploration methods and provides examples of how non-invasive exploration technologies like spectral imaging can help. These technologies allow for rapid, multi-scale mapping of material characteristics in a way that is non-invasive, versatile and can provide automated real-time processing. The document also discusses using these methods from drones to map inaccessible areas and for environmental monitoring, as well as the need for multimodal 3D processing and potential for autonomous drone swarms.
Sandra Lorenz WiMLDS Dresden
- 1. Helmholtz-Institute Freiberg for Resource Technology · Department of Exploration · Dr. Sandra Lorenz · s.lorenz@hzdr.de · www.hzdr.de The role of data science and machine learning to achieve sustainable mineral exploration
- 3. Challenges ▪ Cost ▪ Timing ▪ Invasiveness ▪ Acceptance ▪ Accessibility ▪ Safety ▪ Scale Gap CONVENTIONAL MINERAL EXPLORATION Photo: Moritz Kirsch Lorenz 2019  3
- 4. E.g., Spectral Imaging for surface mapping of material characteristics ▪ non-invasive ▪ versatile ▪ rapid ▪ multi-scale ▪ multi-view NON-INVASIVE EXPLORATION TECHNOLOGY Lorenz 2019 Photos: HZDR-HIF  4
- 5. SPECTRAL IMAGING FOR RESOURCE CHARACTERIZATION Lorenz 2019  5
- 6. SPECTRAL IMAGING FOR RESOURCE CHARACTERIZATION Lorenz 2019, Booysen et al. 2021  6
- 8.  8
- 9. 200 mm HYPERSPECTRAL DRILL CORE MAPPING Photos: HZDR-HIF
- 11. CLASSIFICATION Lorenz et al., 2021  11
- 15.  15
- 16. DIGITAL OUTCROPS AND OPEN MINE PIT MAPPING Photos: M. Kirsch  16
- 17. 2D image space vs. 3D point cloud space THE NEED FOR MULTIMODAL 3D PROCESSING 3D MULTIMODAL POINT CLOUD The world is 3D, so AI should see it in 3D too! ❖ major advances in image processing and AI (feature extraction, denoising, classification, ...) ❖ distorted spatial relations ❖ occlusions ❖ joining spatial features (shape, texture) with multimodal data characteristics (spectra, ..) ❖ object detection and segmentation ❖ limited AI implementation for monomodal point clouds (3D- CNN, PointNet) 2D SPECTRAL IMAGING
- 18. Thiele et al. 2021 (under rev.) CLASSIFICATION OF MULTIMODAL 3D DATA
- 19. DRONE-BORNE HIGH- RESOLUTION MAPPING OF REMOTE, INACCESSIBLE AREAS Jakob et al., 2017 Photos: HZDR-HIF
- 20. (offset for clarity) wavelength (nm) raw corrected with topographic correction uncorrected GEOMETRIC CORRECTION RADIOMETRIC CORRECTION COMPLEX DATA CORRECTIONS Lorenz 2019  20
- 21. DRONE-BORNE HIGH-RESOLUTION MAPPING OF RARE EARTH ELEMENTS Booysen et al., 2021
- 22. DRONE-BORNE ENVIRONMENTAL MONITORING e.g., mapping of Acid Mine Drainage Flores et al., 2021 50 mm  22
- 24. ✓ Enhance worker safety by autonomous processes ✓ Reduce mapping costs ✓ Reduce drilling ✓ Enhance mapping accuracy, speed and coverage ✓ Reduce waste and ore processing effort by better preselection ✓ Environmental impact monitoring AI-BOOSTED NON-INVASIVE EXPLORATION TECHNOLOGY  24
- 25. THANK YOU!