Jump to the following:

We use cookies to improve this website. Read about cookies

Microsoft and Ordnance Survey join forces to teach machines how to identify types of roofs

Machine goes from zero to 87% accuracy in five days.

14 February 2018

The hack, featuring software engineers from Microsoft who had travelled from across Europe and Africa to work with OS’s machine learning team, used the city of Hull as a testbed. The trained machine model finished the week by correctly identifying 87% of the roof types it was shown.

In its training the model was shown 500 flat roofs and 500 hipped/gabled roofs, set a confidence limit of 90%, which means it must be 90% confident or more for its assessment to count.

Isabel Sargent, Senior Research and Development Scientist at OS, says: “Thanks to the excellence of the Microsoft team we have been able to work out together how to stream this machine captured data into our database for if and when we’re ready to put machine learning into production. It’s already very accurate, going from zero to 87% accuracy in just one week, but we need to increase its success rate. Although much slower, humans typically have an error rate of around 5%.”

In the time it takes a human to classify roof types in a single image, a machine can process thousands of images. It is thought it would take the machine less than a day to classify all 35,700,655 properties in Great Britain and their roof types. This enhanced level of detail could inform insurance companies to help them produce more accurate risk assessment.

Isabel continues: “We now have a machine training system that can be trained to classify almost anything, such as bodies of water for example, and on top of this, the machine can be set to run hundreds of these classification queries simultaneously.

“While it can take time to teach a machine, once that deep learning is done the classification algorithm is there ready to decode the landscape. This has the possibility to significantly increase the efficiency of our surveying operations – freeing up the surveyors we have on the ground to focus on those complex tasks for which there is no AI assistance. We would not have been able to achieve this breakthrough so quickly without Microsoft, and we’re thrilled for the future by the potential of what our collaboration has achieved.”

“The week-long hack we’ve conducted with Ordnance Survey demonstrates the potential benefits that artificial intelligence can bring to people in a broad mix of roles,” says Gina Dragulin, Audience Evangelist, Microsoft UK.

“Our own research (PDF) recently identified that workers believe that more than a third (36%) of tasks they carry out on a regular basis would benefit from intelligent automation. This hack is a perfect example of the value that could be created for Ordnance Survey surveyors if they were able to eliminate time-intensive tasks, such as roof-type identification, from their day-to-day activities, enabling them to focus on more complex tasks that require the type of high-level decision making and judgement calls that only an expert can make.”

This work follows on from a project for the Rural Payments Agency that saw OS utilise machine learning to automate the identification and mapping of 373,919km of England’s farmland hedges. The project produced consistent results and proved more cost effective and efficient than manual data capture. To help the UAE better manage Climate Change and its natural resources, OS has developed a machine learning process to identify palm trees from satellite imagery.

ENDS

View video in full

Contact us

For more information:

Contact: Keegan Wilson, Senior Press Officer

Email: keegan.wilson@os.uk

Phone: (+44) 023­80 055332

Latest news

Seven start-ups left the inaugural Geovation Conference with a share of £155,000 in funding after impressing judges at the Royal Geographical Society in London.

A week-long machine learning hack took place at Ordnance Survey (OS) Head Office to train machines to classify roof types contained within the national mapping agency’s aerial imagery.

Back to top
© Ordnance Survey 2016