Geology Machine Learning
Machine learning in geology holds a great deal of promise as a tool for improving the efficiency and quality of geological investigation, interpretation and modelling.
While machine learning is evolving quickly, it is still at the early stages in the technology cycle and will mature steadily to deliver exceptional value for exploration and mining.
Imago is working with companies around the globe to develop machine learning using geology images, including improving the photography process and workflows to use the outcomes of machine learning.
While AI is a broad concept meaning machine intelligence (emulating human intelligence), machine learning is the concept of training a computer to classify, count or predict something based on data inputs. Advances in the quality of images with low cost cameras, together with Cloud computing have led to the opportunity to use drilling core and chip images to help geologists make faster, more accurate decisions.
The journey to machine learning is a collaborative process involving senior geologists, data scientists and IT team. It is an iterative process where geologists define and supervise the problem to solve, data scientists focus on the data and training the model, and IT team ensures results can be operationized to deliver the outcomes to the geology end users.
The first steps towards machine learning are:
- Improve the quality and consistency of your drilling core and chip photographs to ensure they are fit for machine learning. (Zhang, Wang, Li, Han 2018)
- Build a library of images that makes images instantly available to geologists during interpretation and modeling. This then leads to collaborative problem identification that becomes the focus of the first machine learning experiments.
- Having your images stored and viewable using Imago makes collaboration with internal stakeholders and machine learning consultants easier as they review images, explore potential problems and review learnings from each iteration.
- Having your images in the Imago Cloud library also means efficiency in sending images to Cloud based machine learning models and delivering results to the end users.
This image shows the concept of clicking a drill hole trace (Leapfrog) – the image opens in a browser, viewed as a drill hole, allowing the geologist to validate their interpretation by viewing the core or chip images as an integral part of their daily work. Machine learning has the potential provide additional insights to your team.
Machine Learning using Geological Photos
The stages to realize the value of machine learning are:
- Establish a consistent process for capturing high quality images, stored in a modern cloud based geological image library.
- Integrate the image library with your preferred geological and mining tools, so that:
- Geologists can use the images to validate their daily work.
- Consultants can review image datasets from any location to access potential machine learning projects based on the library of images.
Integrating high quality images with existing workflows will help the team understand the current value and further potential value from high quality images and machine learning.
Developing machine learning models is a specialized activity that will require an investment in research. Justifying that research will require a clear value proposition with two components:
- The problem or opportunity the business aims to solve with machine learning. Examples could be:
- Identify instances of poor classification of critical lithological units during human logging.
- Determine the % of mineral types that are visually distinctive, such as sulfides.
- Particle size analysis of plant feed.
- Geomechanical properties from images.
- Structures, RQD.
- Image quality.
- The potential business value of solving the problem or realizing the opportunity.
Steps to Machine Learning using Geological Images
After identifying a problem or opportunity and assessing the value, you may wish to embark on a machine learning project. The steps to developing and implementing machine learning are as follows:
Establish an efficient workflow for capturing consistent, high quality photographs, stored in a fit for ML cloud library.
Integrate the image library with the daily tools and tasks of the geology team. This helps the team understand both the process and the incredible value of getting the basics right for core and chip tray image capture and instant access for geology and drilling supervision. It also provides the image viewing tools that are needed for collaboration of participants throughout the project.
Clarify the problem/opportunity and value of potential machine learning projects.
Review the available data and determine its data potential for solving the problem or realizing the opportunity.
Consider the process (hardware and workflow) for capturing the required images for the purpose. The process for collecting photos and implementing outcomes must be feasible.
Prepare the images and any other data that will be used to train the machine learning model.
Train the model using the prepared images and test its performance against the objective.
Repeat steps 6 and 7 to improve the performance of the model.
Implement the required operational data capture process.
Implement a system for operationalizing the machine learning process:
- As images are captured, automatically feed images to the machine learning model.
- Retrieve, store machine learning results.
- Make results available to geology or operations teams/decision points.
Review and improve the machine learning model and operational process.
Consider other related applications for the ML model.
Imago role in machine learning using geological photographs
Imago provides an image capture and cloud based geological image library required for steps 1 and 2 of the ML journey above. Imago is integrated with the popular geological and mine planning tools to provide instant viewing of images to support and validate geological decisions.
Imago helps customers implement a sustainable image capture workflow to achieve consistent, high quality images.
Imago provides the tools to mask and classify images for machine learning model training.
Finally, Imago operationalizes machine learning using geology images by defining actions on new images, sending images to a machine learning model, and delivering results to geologists.
- In the core shed or core yard, a Windows Laptop or Tablet is tethered to the camera.
- The Imago Capture software (Desktop) is installed on the Laptop. When online, the user signs into Imago web and settings are automatically synchronized.
- The capture software does not require internet access to capture images.
- Images are taken by clicking a button on the PC using the capture software. A single click takes the photograph, names it and moves the image to the PC. This eliminates the requirement for users to touch the camera or access the camera SD card to retrieve images.
- At the end of a shift, or day, when the PC is online, the images are synced to the Imago cloud library where they are instantly available to the geology team and supervisors.
- If communications at site are poor or intermittent, the PC or mobile storage device can be taken to an area with communications for syncing at the later time.
Accessing and viewing images
Once the images are synced to the Imago Cloud library, they are available for viewing from any location that has internet and Login credentials.
Integration with geology and mine planning tools
Each image is identified by a unique URL. Image URL’s can be exported to the Geology database, or constructed following a standard format from Hole ID, depth from, depth to. The links can be imported to the Geology Database of choice (e.g. acQuire, MX Deposit, Datashed), or imported to a project file/database.
Any image can be opened and viewed as a drillhole by clicked a generic link to the image from within a database or mine planning tool.
Data security and backup
- Imago uses the Microsoft Azure cloud storage which ensure the world’s leading data security.
- Imago data security has passed the standards of some of the mining industries largest organizations.
- Images can be automatically copied to an internal customer file store if required.