Convolutional neural networks allowed us to create a reliable solution and take advantage of recent developments in the hot area of deep learning.
Palo Alto, CA (PRWEB)
March 23, 2017
The Defence Science and Technology Laboratory (Dstl), a UK government agency, recently hosted a competition to identify and categorize objects in satellite imagery. Dstl provided 1 km x 1 km satellite images and the challenging task was to detect different types of objects, such as buildings, vehicles, trees or roads located in various environments. Users had to find an algorithm or develop software that would help evaluate large and complex data sets in an innovative manner.
“Coming up with a model that can generalize well from a small training set of only 25 multispectral images was a really challenging task. Convolutional neural networks allowed us to create a reliable solution and take advantage of recent developments in the hot area of deep learning. We were excited to apply novel scientific solutions for practical problems and see their excellent performance,” said Robert Bogucki, Chief Science Officer at deepsense.io.
deepsense.io’s data science team took fourth place with their creative solution, beating over 400 other teams from all over the world. There were over 5,000 submissions on Kaggle, which translates into over £2.5m worth of research. The results can be used in a great number of projects requiring a quick and accurate analysis of large data sets.
“I am proud of our machine learning team, who proved their data science skills in the Kaggle competition once again. We took such a high place due to the fact that we have outstanding and talented people who use appropriate methodology and tools, such as Neptune, developed…