(50-1) 12 * << * >> * Russian * English * Content * All Issues
DOI: 10.18287/COJ1657
Article ID: 1657
Language: Russian
Abstract:
The article presents the results of an experimental evaluation of the efficiency of machine vi-sion systems based on CMOS matrices with a mask of polarizing filters in comparison with CMOS matrices with a Bayer mask based on wavelength filters in detecting low–contrast objects of inter-est using artificial neural network algorithms under different scene illumination. The possibility of using artificial neural networks with a three–channel input trained on data with channel differen-tiation by wavelength is shown for analyzing data whose differentiation is based on wave polari-zation. The advantage of using data on the polarization of light waves in comparison with data on the wavelength of light in recognizing low–contrast objects of interest in a number of condi-tions is shown. In particular, the use of a polarization vision system allows expanding the operat-ing range of illumination and increasing the reliability of detection by several orders of magnitude in comparison with a color vision system for various objects of interest in the studied illumination range. We are also making publicly available for the first time a dataset of machine vision camer-as based on a CMOS matrix with a mask of polarizing filters and a CMOS matrix with a Bayer mask, obtained in synchronous mode with the same shooting parameters for the same scene with different illumination.
Keywords:
polarization, artificial neural network, image processing, optical information technologies.
Acknowledgements:
This work was supported by the grant of the Russian Science Foundation No. 24–19–00063, “Theoretical foundations and methods of group control of unmanned underwater vehicles”, https://rscf.ru/project/24–19–00063/ on the basis of the Federal State Autonomous Educational Institution of Higher Education “Southern Federal University”.
Citation:
Budko AY. Budko NA, Karaev AD. Recognition of low-contrast objects of interest in low-light conditions based on neural network algorithms and light wave polarization. Computer Optics 2026; 50(1): 1657. DOI: 10.18287/COJ1657.
References:
151, Molodogvardeiskaya str., Samara, 443001, Russia; E-mail: journal@computeroptics.ru; Tel: +7 (846) 242-41-24 (Executive secretary), +7 (846) 332-56-22 (Issuing editor), Fax: +7 (846) 332-56-20