Formation of training and control sets of synthesized images, results of training the YOLO neural network
Аuthors
1*, 2, 21. Closed Joint Stock Company «Technological Park of Cosmonautics «LINKOS»», Moscow, Shcherbinka, Russia
2. State Research Institute of aviation systems, Moscow, Russia
*e-mail: a_krasnov@inbox.ru
Abstract
The formation of training and test image sets includes the following steps: preparation of initial data; energy calculation and synthesis of initial images of objects and background; synthesis of target environment images. Initial data preparation involves: determining data on the observed objects and the underlying surface (background); determining data on the state of the atmosphere (external conditions); defining and calculating the characteristics of the electro-optical system; and defining the target environment and external conditions.
In turn, the result of defining the target environment and external conditions is the initial data required to perform the energy calculation:
- constant initial data, the values of which do not change during the energy calculation;
- variable initial data, the values of which change during the energy calculation.
The specific list of initial data is determined by specialized software designed for performing the energy calculation, the conditions under which the objects are being photographed, and the location of the object relative to the electro-optical system, the characteristics of which will be taken into account during image synthesis.
MODTRAN, a program designed to calculate atmospheric transmittance and optical radiance using several atmospheric models, was used to perform the energy calculation.
The aim of this work is to substantiate and generate initial data whose values are modified during energy calculations using MODTRAN, train the YOLO neural network, and analyze the training results. The initial data includes the atmospheric model, background and object temperatures, object range, position of the extraterrestrial radiation source, meteorological visibility, and an aerosol model. This is a minimal set of variable initial data; it may change depending on the external conditions involved in the image synthesis task.
Keywords:
neural network training and learning control; computer vision model; YOLO; energy calculation; MODTRANReferences
- Krasnov A.M., Shashkov S.N., Rumyantsev A.V. Trudy MAI: elektron. zhurn, 2025. no. 144. Avialable at: https://trudymai.ru/published.php?ID=186316.
- Krasnov A.M., Shashkov S.N., Rumyantsev A.V. Trudy MAI: elektron. zhurn, 2026. no. 146. Avialable at: https://trudymai.ru/published.php?ID=187455.
- Berk A., Anderson G.P., Acharya P.K., Hoke M.L., Chetwynd J.H., Bernstein L.S., Shettle E.P., Matthew M.W., Adler-Golden S.M. MODTRAN4 Version 3 Revision 1 User’s Manual. Technical Report. Air Force Research Laboratory, Hanscom Air Force Base, MA, USA, 2003.
- Ientilucci E.J. Using MODTRAN predicting sensor-reaching radiance. Chester F. Carlson Center for Imaging Science. Rochester Institute of Technology, 2007.
- Driggers R.G., Friedman M.H., Devitt J.W., Furxhi O., Singh A. Introduction to Infrared and electro-optical systems, Third edition, Artech House, 2022.
- Berk A., Anderson G.P., Bernstein L.S., Acharya P.K., Dothe H., Matthew M.W., Adler-Golden S.M., Chetwynd J.H. Jr., Richtsmeier S.C., Pukall B., Allred C.L., Jeong L.S., Hoke M.L. Modtran4 radiative transfer modeling for atmospheric correction. SPIE Proceeding, Optical Spectroscopic Techniques and Instrumentation for Atmospheric and Space Research III, 1999. Vol. 3756. Art. 348. (Proceedings of SPIE – The International Society for Optical Engineering).
- Holst G.C. Electro-optical imaging system performance, Sixth edition, JCD Publishing and SPIE Press, 2017.
- Atmospheric correction module: QUAC and FLAASH User’s Guide. ITT Visual Information Solutions, 2009.
- Krasnov A.M., Shashkov S.N., Rumyantsev A.V. Trudy MAI: elektron. zhurn, 2025. no. 145. Avialable at: https://trudymai.ru/publications.php?ID=186886
- Pogoda i klimat, website. Avialable at: https://www.pogodaiklimat.ru
- Zuev V.E., Banakh V.A., and Pokasov V.V. Sovremennye problemy atmosfernoi optiki. Vol. 5. Optika turbulentnoi atmosfery. Leningrad, Gidrometeoizdat, 1988.
- Krasnov A.M., Tregubenkov S.Yu., Rumyantsev A.V., Khismatov R.F., Shashkov S.N. Informatsionno-izmeritel’nye i upravlyayushchie sistemy, 2021, vol. 19, no. 1. pp. 45–64. DOI 10.18127/j20700814-202101-04
- Krasnov A.M., Tregubenkov S.Yu., Rumyantsev A.V., Khismatov R.F., Shashkov S.N. Trudy MAI: elektron. zhurn, 2022, no.122. DOI 10.34759/trd-2022-122-22
- Suncalc.org: website. URL: https://www.suncalc.org.
- Mooncalc.org: website. URL: https://www.mooncalc.org.
- Gavrilov V.A. Vidimost’ v atmosfere. Leningrad, Gidrometeorologicheskoe izdatel’stvo, 1966. 324 p.
- Shettle E.P., Fenn R.W. Models for the aerosols of the lower atmosphere and the effects of humidity variations on their optical properties. Air Force Geophysics Laboratory, Massachusetts, 1979.
- Ultralytics: website. Avialable at: https://ultralytics.com.
- Driggers R.G., Friedman M.H., Nichols J. Introduction to infrared and electro-optical systems, Second edition, Boston; London, Artech House, 2012.
- Holst G.C. Electro-optical imaging system performanc, Fifth edition, JCD Publishing and SPIE Press, 2008.
Download