A hot topic that is profoundly affecting the geospatial industry is the concept of deep learning and machine learning.
Many organisations are developing methods to apply deep learning, expose algorithms to allow experimentation with classifiers, or trying to gain insights from mining big data.
It's a relatively new way of performing geospatial data analysis and any advantages that can be gained from predictive analysis means less down time, fewer costly repairs, and a faster path to results.
Often, the data we work with doesn't always come from an overhead sensor, such as satellite or aerial data. What if we need to work with data on a surface and we are scanning to the side or vertically overhead? Or both?
The challenge that we accept is how to start analysing this type of data and perform machine learning on it, without a traditional geospatial context.
In countries with advanced imagery capabilities and geospatial technologies, such as Australia, people are asking these difficult questions.
One question that I've seen come up a few times now is related to our machine learning capabilities and applying these techniques to various types of sheet metal, such as ship building yards or wind turbine inspections.
Both of these types of structures are very complex and need experts to appropriately capture, organise, and analyse the imagery.
We have been developing and implementing machine learning solutions to solve real world problems like this, working with all types of data, whether it’s satellite imagery, aerial data, photographs taken from smart phones or photographs from a mounted camera or sensor.
Through this, we bring imagery expertise and deep learning to projects. One example is for wind turbine inspections, where we can discern damage that requires repair, instead of damage that doesn’t need to be repaired, from bird strikes, for example. This saves time and cost for the asset owners, ensuring that essential maintenance is accurately identified and prioritised.
This solution can be repeatedly applied to produce correct damage assessment accuracies of greater than 95%. I see this technology successfully answering the same questions for organisations across Australia – given the land mass, the inaccessibility of some structures, or the sheer physical size of some assets, such as large ships and energy infrastructure.
Through our understanding of deep learning and machine learning, we can apply these techniques and bring solutions to use technology to overcome the challenges of geographic reach and accurate imagery analysis.
About the author
Asia Pacific Regional Manager