01.02.2024, 09:52
Russia has created a high-speed neural network for processing data from UAVs
Source: OREANDA-NEWS
OREANDA-NEWS Samara University scientists have developed a high-speed neural network capable of analyzing hyperspectral data from drones or space satellites, the university's press service reported.
"Scientists of Samara State University named after Korolev has developed and tested a high-speed neural network capable of analyzing the incoming video stream in real time and almost instantly recognizing and finding specified objects and images in this video stream. Along with analyzing the image from a conventional video camera, the development can also quickly analyze data obtained using hyperspectrometers — devices that see reality in multichannel spectral imaging and allow detecting objects invisible to conventional means of observation," the Samara University said in a statement.
The optical neural network was created on the basis of an analog photonic computing system, which has a number of advantages: full protection from electromagnetic interference, low energy consumption and the possibility of parallel data processing. Scientists have created a demonstration sample using standard laboratory optical and mechanical components, various modulators and video cameras, said Roman Skidanov, professor of the Department of Technical Cybernetics at Samara University.
"The recognition reliability during the first experiments on the demonstration sample was 93.75%. In 2024, it is planned to assemble and test an experimental sample of the system in a fairly compact package the size of a small computer system unit. The accuracy and reliability of recognition in the experimental sample should increase due to the selection of components with improved characteristics," Skidanov noted.
With hyperspectral photography or hyperspectral remote sensing of the Earth, which are carried out using drones or a space satellite, it is possible to effectively find greenhouse gases, conduct geological exploration of hard-to-reach areas, as well as more accurately track the occurrence of forest fires.
"Scientists of Samara State University named after Korolev has developed and tested a high-speed neural network capable of analyzing the incoming video stream in real time and almost instantly recognizing and finding specified objects and images in this video stream. Along with analyzing the image from a conventional video camera, the development can also quickly analyze data obtained using hyperspectrometers — devices that see reality in multichannel spectral imaging and allow detecting objects invisible to conventional means of observation," the Samara University said in a statement.
The optical neural network was created on the basis of an analog photonic computing system, which has a number of advantages: full protection from electromagnetic interference, low energy consumption and the possibility of parallel data processing. Scientists have created a demonstration sample using standard laboratory optical and mechanical components, various modulators and video cameras, said Roman Skidanov, professor of the Department of Technical Cybernetics at Samara University.
"The recognition reliability during the first experiments on the demonstration sample was 93.75%. In 2024, it is planned to assemble and test an experimental sample of the system in a fairly compact package the size of a small computer system unit. The accuracy and reliability of recognition in the experimental sample should increase due to the selection of components with improved characteristics," Skidanov noted.
With hyperspectral photography or hyperspectral remote sensing of the Earth, which are carried out using drones or a space satellite, it is possible to effectively find greenhouse gases, conduct geological exploration of hard-to-reach areas, as well as more accurately track the occurrence of forest fires.
Комментарии