New 'Smart Vision' System for UAVs Developed by Armenian and Russian Scientists

Thanks to the developments of Armenian and Russian scientists, it has become possible to implement an automated and affordable 'smart vision' system in satellites and unmanned aerial vehicles (UAVs). According to Armenpress, citing RIA Novosti, this was announced by the press service of the Sergey Korolev National Research University.

The scientists note that this system will facilitate the implementation of hyperspectral technology. Spectrometers are devices that allow obtaining data about the composition of a material by separating numerous individual waves of electromagnetic waves. While a traditional camera captures only three such wavelengths, modern hyperspectrometers can provide information from 100 or more spectral channels. Scientists have reported that digital hyperspectral photographs are referred to as 'hypercubes'.

For the recognition of images recorded in hypercubes, neural networks are currently used, which compare the images with data from sample masses. The preservation and processing of such benchmark mass requires significant resources. As scientists explain, each time different types of objects' hypercube analysis necessitates the selection of key features through comprehensive and in-depth analyses.

A new algorithm jointly created by Armenian and Russian scientists will help overcome these challenges and significantly enhance the speed of 'smart vision' operation. According to them, the algorithm allows replacing the mass of hyperspectral samples with a pre-approved set of signs corresponding to the current task of a drone or satellite.

“Our approach allows for the management of one parameter, selecting an optimal filter for processing the entire image. Based on this, we are developing a self-learning algorithm that can determine the informational signs of required objects in hypercubes without the need to collect samples manually. Our proposed solution will enable the creation of mobile hyperspectrometers that will recognize the necessary objects as early as this summer,” said project leader, Professor Alexander Kupriianov, head of the Department of Technical Cybernetics at Samara University.

Currently, the use of hyperspectral 'vision' in nanosatellites and lightweight UAVs is not economically viable as it requires the storage of a large volume of collected data. Moreover, transferring this data to a server for neural network processing in field conditions is even more challenging as it requires substantial time and a wide bandwidth. With the implementation of the new algorithm, scientists are confident that hyperspectral vision may operate in offline mode.

“We plan to release a prototype of a universal computing system in 2022 that adapts to any problem of image analysis through the automatic selection of specific informational signs. Such a system will significantly increase the efficiency of solving various applied issues related to digital image analysis, including geoinformatics, 'smart' agriculture, remote sensing of the Earth, and even in medical diagnostics,” noted Kupriianov.

The project has received grant support from the Ministry of Education, Science, Culture and Sports of the Republic of Armenia and was conducted in collaboration with specialists from the Armenian-Russian University.