GoldSpot’s LithoLens, a proprietary core imaging technology, automatically and consistently examines old core images to create fresh and accurate geological logs. This proprietary tool turns old core photos into intact, georeferenced core images and uses deep learning algorithms to enhance the images and extract valuable geological information.
The software platform autonomously extracts information from drill core images and video, outcrop photography, and within-borehole sensors. Secure cloud-based data processing and machine learning creates large and valuable new datasets from underutilized imagery and video data.
Drillcore Photo Analytics
Every mineral exploration company is legally required to photograph drillcore as it is collected. These drillcores are later visually inspected by geologists, and characteristics of the core, such as lithology, alteration, and mineralization, are manually recorded in composite intervals.
Most deposits have accumulated – and continue to collect – thousands of core photos, most often stored on various servers and forgotten. Millions of dollars may have been spent collecting this overlooked information.
We extract untapped value from those existing core box photos by using deep learning to eliminate aspects of the image that are not rock, and clean and optimize the quality of the existing images. We then employ LithoLens to automatically recognize varying geological intervals – or specific features such as veins – within the core. Our process allows us to utilize existing images and extract relevant information that might have been missed through the core logging process.
Downhole Optical Analytics
Downhole camera technology can be lowered into empty drillhole to capture a 360º image of the rock along its entire depth. Interpretation of these photographs has traditionally required time-consuming manual cataloguing and analysis using specialized and expensive software.
Downhole Optical Analytics automates the assembly and analysis of imagery collected by drilling contractors. In addition to quicker and more accurate reconstruction of the entire drillhole image, our LithoLens is used to split the final image into segments of interest, and then categorize the image to instruct our deep learning module to automatically locate similar information from the remaining dataset.
This same approach is used to quickly and efficiently analyze underground rock headings in existing mines.
Satellite Image Analysis
To refine broad exploration areas to more manageable target sizes, GoldSpot obtains and processes satellite imagery – such as LandSat, ASTER, HyMap, or PlanetScope – and create geological maps. Machine-learning can then be utilized to support exploration efforts such as:
- Mapping geological zones
- Normalizing regional geochemical sampling to allow ranking of metal anomalies
- Interpreting mineralization showings as likely to be in situ vs. transported from source
- Planning of field work
GoldSpot routinely processes and interprets airborne or satellite multispectral imagery – such as Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) – to differentiate alteration assemblages and vector towards hydrothermal mineral deposits. Complex and overprinting alteration patterns can be unravelled with the VNIR, SWIR, or TIR-active mineralogy, as well as changes in mineral chemistry or composition, all of which help broaden understanding of hydrothermal systems and ultimately guide the explorer more directly to the core of the system.
In heavily forested areas it can be difficult and time consuming to find physical rock outcroppings necessary for accurate geological mapping or sample collection. Traditionally, sites with good outcrop likelihood have been selected by manually studying aerial photos or satellite images and manually recording coordinates.
Our geologists train Deep Learning algorithms to determine areas of outcrop. By automating the process, weeks of physical image review are reduced to a matter of hours. Furthermore, geologists in the office are able to relay new potential targets directly to mapping tablets in the field. Initial development in Northern Ontario and Northern Quebec has shown considerable success, and we are continuing to broaden our capabilities in other geological regions.
Undersea Image Recognition
One of the newest opportunities in mining is the collection of manganese nodules from the ocean floor. These naturally occurring rock concretions are found in high concentrations on the seabed in several areas around the world, often at depths of between 4,000 to 6,000 meters.
To improve the efficiency of their retrieval from the sea floor, we have developed a resource estimation process to determine the areas of
greatest potential for undersea mining. Using camera images of the sea floor and the few samples brought to the surface for analysis, we can quickly and accurately estimate the total amount of resources available in each area, providing precise modelling. In the future, we will be able to use LithoLens to process images from cameras attached to the undersea mining vehicles themselves, to make mineral content and economic assessments in real time.