UDC 004.932
FILTERING NOISY IMAGES BASED ON NEURAL NETWORK PROCESSING OF SPATIALLY ORIENTED WAVELET TRANSFORM TREES
Y. S. Bekhtin, Doctor of Technical Sciences, Professor, Department of Automatics and Information Technologies in Control, RSREU, Ryazan, Russia;
orcid.org/ 0000-0001-9963-7244, e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
V. T. Trinh, post-graduate student, RSREU, Ryazan, Russia;
orcid.org/ 0009-0007-3697-4797, e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
This article addresses the pressing problem of noise reduction in synthetic aperture radar (SAR) images. The focus is on effectively suppressing speckle noise, characteristic of SAR data, while preserving important image details such as object boundaries and texture features. A new filtering method based on the use of spa tially-oriented trees (SOTs) of wavelet transform is proposed. The method employs a recurrent neural net work with long short-term memory architecture applied to coefficient sequences extracted from SOT struc ture to more accurately model the dependencies between decomposition levels and improve the quality of noisy image reconstruction. Experimental studies conducted on the synthesized noisy SAR images demon strate the superiority of the proposed approach over traditional filtering methods in both visual quality of the results and objective metrics.
Key words: : SAR image, wavelet transform, speckle noise, spatially-oriented trees, recurrent neural net works, long short-term memory.