Optimising core tray imagery

A MINING software maker has applied machine learning to optimise value in historical core tray imagery.
Optimising core tray imagery Optimising core tray imagery Optimising core tray imagery Optimising core tray imagery Optimising core tray imagery

Micromine has developed a way to get machines to take a better look at core photos.

Micromine's Perth office recently took part in The Newcrest Crowd, a crowdsourcing and partnership platform run by Unearthed, to solve specific mining problems through online competitions.

The "Get 2 the Core" competition focused on core tray photography and how companies could get value from historical core tray imagery.

While Micromine did not win - that honour went to Straight off the Couch - it has gained a handy tool that it will be able to apply to its Geobank data management solution.

Some believe core tray photography is not as widely used as it could be due to the arduous process of capturing, processing and analysing core tray photography.

With so much rich textural, mineralogical and geotechnical information contained in core tray photographs there is value in optimising the process.

Photographs taken on Newcrest sites are mostly standardised, however, they have millions of historic core images from tens of thousands of drill holes.

As technology evolves, image analysis techniques are becoming more powerful and prevalent within the mining industry.

These large image repositories should eventually become rich sources of quantitative data.

"Get 2 the Core" asked participants to build an algorithm to determine and map the spatial extents of the core tray and then the individual rows contained within.

The successful participant was awarded a $10,000 prize, with a separate prize awarded for solutions that also solved the problem but did not fit the scoring requirements.

Micromine's Wojciech Slabik said the Micromine Pitram team had been working with machine learning techniques to solve mining problems so it noticed it could apply those methods to the Newcrest problem.

It applied machine learning skills using the relatively new Mask R-CNN technique.

"We utilised a technique known as Transfer Learning so our Mas R-CNN learned very quickly to deal with the core tray data," Slabik said.

"Using Transfer Learning meant we didn't need thousands of labelled core trays to be able to train the machine.

"To generate the core bounding box we then used more traditional edge detection techniques that used the mask from the Mask R-CNN."

Participants were given a training dataset of images and completed masking instructions.

The test data set consisted of images only, for which the participants needed to predict the masking instructions via a comma separated values file.

The solution needed to be able to perform on inconsistent photograph where:

  • Core boxes can be made from different materials such as wood, steel, cardboard or plastic;
  • Images are highly variable in terms of resolution, aspect ratio and quality; and
  • The relative position of the core tray within the image can be variable.

The competition aimed to reduce the man hours needed to manually review and analyse core tray photography.

Due to the manual mark up the labels may not be perfectly consistent across the entire dataset, however, these labels represent the kind of work needed to produce an outcome that is time consuming and labour intensive.

"The results we got in only a few hours were much better than we initially expected," Slabik said.

"Future development will be focused on training the network to generate not just the bit mask outlining the drill core but also the full bounding box."