Soologic Technological Solutions S.L is a technology-based company with a strong involvement in R&D projects. The company has participated in the European H2020 project “Robotics for Infrastructure Inspection and Maintenance (RIMA)”. Within this project, the research group has been contracted for image preprocessing and the development of machine learning algorithms, in particular deep learning.
The objective of this project is to design a system for the detection and classification of airport pavement damage using an aerial drone, specifically designed and equipped with a high resolution camera, automatic flight systems and an artificial lighting system for night flights. This drone has been designed, developed and built by the company Flying Robotic Solutions S.L.
The fundamental techniques used are deep learning and transfer learning. The deep learning algorithm is able to learn from other previously classified images, so that after receiving a new image, it is able to determine with high accuracy what type of damage it most resembles. In addition, as flight time within an airport is limited due to heavy traffic, there are often not enough images available to learn, so other sources of images have been incorporated to feed the system and allow it to learn by other means such as transfer learning.
One of the main benefits of using deep learning and transfer learning techniques is the possibility of performing reconnaissance flights at any time of the day or night, without the need for an operator to travel the runways of airports, thus avoiding the safety problems associated with this type of inspections, as well as having a system capable of classifying with very high accuracy and, in addition, improving its performance as more images are taken.