Radiotekhnika
Publishing house Radiotekhnika

"Publishing house Radiotekhnika":
scientific and technical literature.
Books and journals of publishing houses: IPRZHR, RS-PRESS, SCIENCE-PRESS


Тел.: +7 (495) 625-9241

 

Application of convolutional neural networks for thematic mapping of time sequences of satellite multispectral images

Keywords:

V.E. Dementiev – Ph.D.(Eng.), Associate Professor, Department «Telecommunications», Ulyanovsk State Technical University
E-mail: dve@ulntc.ru, vitawed@mail.ru
D.S. Kondratiev – Post-graduate Student, Department «Telecommunications», Ulyanovsk State Technical University
E-mail: kondratev.dmitriy@gmail.com
A.G. Frankel – Post-graduate Student, Department «Telecommunications», Ulyanovsk State Technical University
E-mail: j.skvoll@gmail.com


One of the fundamental steps in the processing of remote sensing imagery is segmentation. The use of known segmentation algorithms for thematic mapping of satellite imagery leads to significant errors due to two reasons. Firstly, the known algorithms are mostly unable to take into account the fact that the satellite image is a multi-zone image. Secondly, the existing approaches do not allow processing the time sequence of such images. The aim of this paper is to overcome these disadvantages by modifying neural network segmentation procedures and classifying multidimensional data.
In this paper, we consider a modification of the convolutional UNET network, based on co-processing and the current multizone image and archival processing results obtained in previous moments. It is shown that this modification allows to improve the quality of treatment by 10−12% and in a significant number of cases to achieve near-pixel accuracy. Compared to the existing algorithms, the proposed approach requires relatively small computational costs and can be implemented in standard image processing packages.

References:
  1. Gonsales R., Vuds R. Czifrovaya obrabotka izobrazhenij. Izd. 3-e. M.: Texnosfera. 2012. 1104 s.
  2. Vasil’ev K.K., Krasheninnikov V.R. Statisticheskij analiz izobrazhenij. Ul’yanovsk: UlGTU. 2014. 214 s.
  3. Fursov V.A., Bibikov S.A., Bajda O.A. Tematicheskaya klassifikacziya giperspektral’ny’x izobrazhenij po pokazatelyu sopryazhennosti // Komp’yuternaya optika. 2014. T. 38. № 1. S. 154−158.
  4. Zimichev E.A., Kazanskij N.L., Serafimovich P.G. Prostranstvennaya klassifikacziya giperspektral’ny’x izobrazhenij s ispol’zovaniem metoda klasterizaczii k-means++ // Komp’yuternaya optika. 2014. T. 38. № 2. S. 281−286.
  5. Tarabalka Y., Benediktsson J.A., Chanussot J. Spectral–spatial classification of hyperspectral imagery based on partitional clustering techniques // IEEE Transactions on Geoscience and Remote Sensing. 2009. V. 47(8). P. 2973−2987.
  6. Dement’ev V.E., Kondrat’ev D.S. Metod tematicheskogo kartografirovaniya posledovatel’nostej sputnikovy’x izobrazhenij // Informaczionno-izmeritel’ny’e i upravlyayushhie sistemy’. 2017. T. 15. № 12. S. 49−53.
  7. Dstl Satellite Imagery Feature Detection. URL = https://www.kaggle.com/c/dstl-satellite-imagery-feature-detection.
June 24, 2020
May 29, 2020

© Издательство «РАДИОТЕХНИКА», 2004-2017            Тел.: (495) 625-9241                   Designed by [SWAP]Studio