R&D project award for geospatial problem solving with Artificial Intelligence

DIELMO D S.L. has been awarded with the R&D project “Analysis of the application of Deep Learning for the resolution of geospatial problems related to territory” and is carrying it out between 2019 and 2022.

This project is part of the call for the Promotion of Digital Enabling Technologies and is funded by the Ministry of Economic Affairs and Digital Transformation of the Government of Spain. The main objective of the project is to study and integrate the currently available technologies of Artificial Intelligence in the resolution of geospatial problems with the territory.

Thanks to the evolution of AI technology, together with the availability of GPUs that enable the parallel processing of data in a fast, economic and powerful way, many applications have emerged in different fields.

This project has focused on deep learning, as it is considered to have a wide field of research applied to geospatial data.

This will require not only integrating AI models and geospatial data in raster or vector format, but also locating the data (and in some cases transforming it into a format that allows working with it), designing the neural network or various configurations of them, implementing an algorithm designed ad-hoc and finally testing with real data to validate its operation and establish the minimum acceptable degree of accuracy.

DIELMO sets high quality levels and being aware of the technological challenge they face, has selected three scenarios:

• Safety of the electrical system, to estimate the horizontal growth of the vegetation from LiDAR data and IFN data and to plan when the existing vegetation will exceed the distance established by regulations, through the use of perceptrons.

• Object detection with geospatial data, in raster and vector format, to be executed through the internet in an online geospatial viewer. In this case, convolutional neural networks (CNN) will be used to detect buildings in orthophotos (using object detection techniques and masks). CNNs will also be used to detect electric towers in high resolution images and for the inventory of the different elements of the electric towers.

• Detection of the exact position of the electric towers and automatic classification of LiDAR point clouds, using 3D convolutional neural networks.

To address the implementation of the proposed activities, we will use open source tools such as Google’s Tensor Flow and will focus resources on designing the neural networks and their configurations, develop ad-hoc algorithms to adapt LiDAR data, raster, vector, orthophotos and high resolution photographs to the available tools and train with large amounts of data so that the system is able to detect what is described in the proposed activities. The challenge is to integrate it into a GIS environment or through the Internet to make it easy to use for non-GIS expert AI users.

The purpose of the R&D project is to integrate software tools, all in free software that allow to implement artificial intelligence models for the treatment of LiDAR data for the resolution of cartographic production problems in which DIELMO is an expert.

We have described three improvable scenarios in which the integration of AI tools, together with the design and development of algorithms necessary for the treatment of input and output data can be effective and allow to offer a robust automatic response.

Input data types such as lidar, raster or vectorial data, orthophotos and high resolution images, common in the daily activity of any entity dedicated to cartographic production and performing revision and maintenance tasks of electrical networks, have been taken into consideration.

As a final result of the project, the internal development team of DIELMO will obtain the knowledge and the necessary tools to solve complex problems that until now we have not been able to solve and to create a base of internal solutions that serve us to optimize our cartographic productive processes and as a base of knowledge that we can use in a future to evolve our department of software development and to offer a new line of services of development of geospatial algorithms supported in Deep learning for final clients at world-wide level like electrical distributors, public organisms (city councils, cadastre, infrastructures), other private companies, etc.

Project financed by the Ministry of Economic Affairs and Digital Transformation.

Call: Promotion of Digital Enabling Technologies.
Beneficiary: Dielmo 3D, S.L.
File number: TSI- 100909-2019-47
Project: Analysis of the application of Deep Learning for the resolution of geospatial problems related to the territory.