Drilling is fundamental to understanding the hidden geology beneath our feet during any exploration, resource definition or mining program. We all love a core image, they provide a long lasting record of the structural, textural and contextual information of the geology. Out of the ~60M metres of drilling undertaken each year, however, roughly 50% of that uses percussive drilling methods (and that’s not including blast holes!). That’s a lot of chip samples that probably aren’t being used to their full capacity.
At Datarock, we have built a Platform that can extract high quality, consistent data from traditional core photography using machine learning. We are always looking to innovate and come up with new ways to extract valuable information from other types of imagery. To that end, we are developing workflows that use computer vision to extract as much information as possible from chip tray images.
This blog is the first in a series talking about what we can do with the humble chip tray image. Like all good science projects, this starts with capturing the best possible data, in this case an RGB image, to give us the best chance of generating robust, quantitative information to help solve your geological problems.
Common issues with historical core photos
One of the key things that can destroy value in any image-based analytics workflow is poor image quality and mistakes made in setting up the sample for imaging. This is particularly the case for chip sample photography where there are so many variables in the type and shape of chip trays and differences in how the photos are taken. This can include things like the number of trays in an image, the lighting conditions, if the chips are wet or dry, washed or unwashed, the list goes on.
The opportunity now exists to utilise modern, ML-based workflows to level up the information you can get out of your chip samples and imagery! In this blog post we will give you some practical tips on how to set up and take the perfect chip tray photo, that can maximise the upside for future machine learning and automation.
Examples of the various types of chip tray type and images commonly available, including some of the problems encountered in these images that make using them for ML problematic.
Setting up for the perfect chip photograph
Similarly to taking a good core tray photo, a big part of getting the perfect chip photo is in the set up and preparation before you take the image. Think of the 3 C’s when setting up your chip tray; Curated, Consistent and Clear!
Tips for setting up your chip tray correctly
- Ensure compartments are filled sequentially in depth order. This includes leaving empty compartments where there are no returns.
- Ensure the depth intervals are consistent throughout a hole, preferably throughout a project.
- Ideally, one tray per image. If you need to include more than one tray, ensure it’s not more than three, so the image is consistently in focus and not too far away. Make sure the trays are in depth order downhole top to bottom.
- A physical guide mark will help ensure the tray is in the correct, consistent position for each image.
- Preferably, have the chip trays in a horizontal orientation, but if not then remember, consistency is key.
- Make sure the tray(s) are on a flat, level surface. The background should be non-reflective, clean and a consistent colour.
- Write the from-to or end depths and the hole id legibly on the inner lid of the tray. This allows us to check digital metadata and ensure the image is in the correct order for analysis
- Curate the chips inside the tray. Make sure each compartment is filled as much as possible and that there is no mixing between compartments. Also make sure that the lid is clean and there’s no mud obscuring the writing there.
- If the chips are wet, ensure there is no ‘standing water’ in the compartment and wipe the lid to make sure there are no droplets there.
- A colour calibration chart should be used in the image. A colour calibration chart can help to track any changes that might occur in the photographic process and also allow for colours to be adjusted/corrected in post processing if required. Colour is one of the key pieces of information in chip samples
In addition to the physical chip tray set up and presentation, other important aspects include the collection of metadata in a digital format. This is really important, as trying to use OCR methods on hand-written numbers on tray lids is difficult and inconsistent.
- Metadata about the image including HoleID and from-to depth information needs to be included in the metadata, either in the image file name, or separate metadata file with a common key to the image file, for example RC001_20-40m_washed.jpg.
- Or, use of machine readable barcodes or QR codes with the metadata embedded in the image.
Example of a ‘best practice’ image setup.
Chip tray camera systems
Given the different scale of the sample compared to drill core imagery, several chip tray camera systems have been developed. These systems typically image either an individual compartment or a set of compartments rather than the entire chip tray. They have a much closer focal length and offer several benefits over using a traditional core photography set up:
- Chip tray cameras produce a higher resolution images (<50 micron compared with 100+ micron with core photography set ups) which allows for more detailed image analysis
- Compartments are in a fixed position, so are easily extracted from the original image
- Compartment imagery will generally have digital metadata for depth assigned during the imaging process.
The overall quality of the image will denote how detailed the analysis can be. This will be explored in our next blog on chip tray analytics.
Examples of different chip compartment image quality. A – Modern high resolution (40 micron) chip tray camera image, with washed/wet chips. B – Moderate quality (100-120 micron) full tray image taken in the sun with shadows. C – Low quality 250+ micron) historical chip image with low lighting.
DO’s and DON’Ts of taking a high-quality chip photograph
- Lighting for the photograph should come from a controlled source inside a light blocking shroud.
- Lighting should be diffuse and high angle to the tray to reduce reflections from wet chips, and to minimise shadows in the compartments.
- Set up the camera with the correct settings and do not change them. Also, do not use Automatic – this will result in variable settings for each photo, ruining consistency. Settings include:
- Autofocus: On (ensure edges of the image are in focus)
- Auto shutter: On
- White balance: Set to match the lighting source used
- Exposure metering: Multi-zone metering (uses many points across the core to minimise dark or light patches)
- ISO: 200-400 is preferred. Set as low as possible to eliminate “grainy” photos (dark rocks may require slightly higher settings)
- Aperture: use a small aperture (< f/7) to ensure a deep field of view
- Lens/image stabilisation: Off (this reduces image distortion)
- Self-timer: 2 seconds (this helps eliminate camera movement)
- Ideally, one chip tray per photo – move the camera close to the tray to fill the image and get the best resolution possible. Make sure the entire tray is captured and in focus.
- Use a modern DSLR with 30-60 Megapixels. This should result in a chip photo with resolution between 100-150 Microns.
- Ensure consistent camera settings and clean lens regularly with a lens cleaning kit.
- Use a light source with the highest possible colour rendering index (CRI).
- Dull coloured non reflective chip trays are best.
- Camera should be held directly above the tray in a fixed frame.
- Taking photos outside using a Sun as the light source
- Illuminating the chips from directly above without a light diffuser – this may result in strong reflections
- Illuminating the chips from a low angle – this may result in shadows in the compartments
- Using the camera flash, particularly without a light diffuser. This will result in reflections on wet samples
- Core shed fluorescent lighting – commonly has a low CRI index
- Reflective backgrounds will cause reflections and potentially over expose the image
Stay tuned for our next blog in our “chip imagery” series. We will be covering how we can analyse chip imagery using computer vision to extract colour, texture and grain size information and convert this imagery into a valuable quantitative downhole dataset.