Hybrid Approach to Agricultural Land Segmentation on Radar Images Integrating Despeckling and Deep Learning

Vladimir Khryaschev; Roman Larionov; Andrey L. Priorov

doi:10.30533/GiA-2026-025

VAK Russia 1.6.19 UDC 528.77 CSCSTI 36.23

3 more Collapse

HYBRID APPROACH TO AGRICULTURAL LAND SEGMENTATION ON RADAR IMAGES INTEGRATING DESPECKLING AND DEEP LEARNING Hybrid Approach to Agricultural Land Segmentation on Radar Images Integrating Despeckling and Deep Learning

АА Аа

Published in Izvestia Vuzov. Geodesy and Aerophotosurveying · Volume 70, Issue 3, 2026 · ELocator: 3 · Rubric: Earth aerospace survey, photogrammetry

DOI: https://doi.org/10.30533/GiA-2026-025

Received: 01.04.2026 Accepted: 24.09.2026 Published: 10.11.2026 Language of publication: ru-RU

Authors

Vladimir Khryaschev

Roman Larionov

Andrey L. Priorov

Vladimir Khryaschev

P.G. Demidov Yaroslavl State University Yaroslavl

Roman Larionov

P.G. Demidov Yaroslavl State University Yaroslavl

Andrey L. Priorov

P.G. Demidov Yaroslavl State University Yaroslavl

ABSTRACT

The problem of agricultural land segmentation on radar images (SAR) is investigated, which, unlike optical images, allow data to be obtained regardless of weather conditions and time of day, but are subject to speckle noise, which significantly complicates automatic analysis. A comparative study of three speckle noise filtering algorithms has been conducted. Evaluation on panchromatic images with simulated multiplicative noise using the PSNR, ENL, and SMPI metrics showed that the BM3D filter provides the best balance between noise suppression and preservation of textural details, which is confirmed by the maximum increase in the PSNR metric and a decrease in the SMPI metric. Based on the collected data set from the Sentinel-1 SAR satellite, experiments on farmland segmentation using modern neural network architectures SegNeXt-L and AerialFormer-B. The use of BM3D pre-filtering allowed us to slightly improve the quality metrics (Dice and F1), as well as accelerate the convergence of neural network models. Further modification of the AerialFormer-B architecture by separating encoders for each polarization channel (VV and VH) provided more efficient feature extraction, which increased the Dice metric to 0.70 and F1 to 0.46.

KEYWORDS

aerospace imagery, Sentinel-1, image filtering, speckle noise, neural network, agricultural land segmentation

1. Baranova DV, Belousov AO, Pavlova VA. Avtomatizatsiya raschetov integral'noi otsenki ratsional'nogo ispol'zovaniya zemel' sel'skokhozyaistvennogo naznacheniya (na primere Novgorodskoi oblasti) [Automation of Calculations of the Integrated Assessment of the Rational Use of Agricultural Land (Using the Example of the Novgorod Region)]. Izvestia Vuzov. Geodesy and Aerophotosurveying. 2025;69(3): 122–135. (In Russian). DOIhttps://doi.org/10.30533/GiA-2025-031. https://miigaik.ru/journal/archive/2025/2025_69_3_RU/GiA-2025-031.pdf

2. Lupian EA, Proshin AA, Burtsev MA, et al. Sistema “VEGA-Science”: vozmozhnosti raboty s dannymi sputnikovykh nablyudenii Zemli dlya resheniya nauchnykh i prikladnykh zadach [VEGA-Science System: Capabilities for Working with Earth Observation Satellite Data to Address Scientific and Applied Problems]. Earth & Universe. 2025;1-2(361-362): 66–85. (In Russian). DOIhttps://doi.org/10.7868/S0044394825010050. https://doi.org/10.7868/S0044394825010050

3. Tetteh G, Gocht A, Erasmi S, et al. Evaluation of Sentinel-1 and Sentinel-2 Feature Sets for Delineating Agricultural Fields in Heterogeneous Landscapes. IEEE Access. 2021;9: 116702 116719. DOIhttps://doi.org/10.1109/ACCESS.2021.3105903. https://literatur.thuenen.de/digbib_extern/dn063902.pdf

4. Baldina EA, Mikhaylyukova PG, Troshko KA. Opyt ispol'zovaniya dannykh radiolokatsionnoi kosmicheskoi s"emki dlya sozdaniya tematicheskikh kart [Experience of Using Radar Satellite Imagery Data to Create Thematic Maps]. Moscow University Bulletin. Series 5, Geography. 2019;4: 110–118. (In Russian).

5. Minaev AS, Fedorov VV. Obratimaya mnogoshagovaya diffuzionnaya model' dlya preobrazovaniya izobrazhenii na osnove radiolokatsionnykh i opticheskikh dannykh [Reversible Multi-Step Diffusion Model for Image Translation Between Radar and Optical Data]. Vestnik Natsional'nogo Issledovatel'skogo Yadernogo Universiteta “MIFI”. 2025;14(3): 249–255. (In Russian). DOIhttps://doi.org/10.26583/vestnik.2025.3.7. https://doi.org/10.26583/vestnik.2025.3.7

6. Shiryaev MA, Baldina EA, Lurie IK. Ispol'zovanie interferometricheskikh dannykh so sputnika Sentinel-1A dlya vyyavleniya izmenenii vysoty poverkhnosti Uzon-Geizernoi vulkano-tektonicheskoi depressii [Use of the Sentinel-1A Satellite DInSAR Data for Surface Elevation Changes Detection of the Uzon-Geyservolcano-Tectonic Depression]. Izvestia Vuzov. Geodesy and Aerophotosurveying. 2024;68(3): 42–56. (In Russian). DOIhttps://doi.org/10.30533/GiA-2024-022. https://miigaik.ru/journal/archive/2024/2024_68_3_RU/GiA-2024-022.pdf

7. Egoshkin NA, Eremeev VV, Kozlov EP, et al. Povyshenie informativnosti kosmicheskikh izobrazhenii Zemli [Increasing the Information Content of Satellite Images of the Earth] [monograph]. Eremeev VV. (ed.). Moscow: KURS; 2021. 368 p. (In Russian).

8. Nafieva EN, Grechishchev AV. Kosmicheskie radiolokatsionnye sistemy monitoringa Zemli [Space Radar Systems of Earth Monitoring]. Ecology Economy Informatics. Geoinformation Technologies and Space Monitoring. 2020;2(5): 89–95. (In Russian). DOIhttps://doi.org/10.23885/2500-123X-2020-2-5-89-95. https://doi.org/10.23885/2500-123X-2020-2-5-89-95

9. Lopez-Martinez C, Fabregas X, Pottier E. Multidimensional Speckle Noise Model. EURASIP Journal on Advances in Signal Processing. 2005;2005: 180956. DOIhttps://doi.org/10.1155/ASP.2005.3259. https://doi.org/10.1155/ASP.2005.3259

10. Dabov K, Foi A, Katkovnik V, et al. Image Denoising by Sparse 3-D Transform-Domain Collaborative Filtering. IEEE Transactions on Image Processing. 2007;16(8): 2080–2095. DOIhttps://doi.org/10.1109/TIP.2007.901238. https://www.sci-hub.ru/10.1109/tip.2007.901238

11. Perera M, Bandara W, Valanarasu J, et al. Transformer-Based SAR Image Despeckling. IGARSS 2022–2022 IEEE International Geoscience and Remote Sensing Symposium. Kuala Lumpur: IEEE; 2022: 751 754. DOIhttps://doi.org/10.1109/IGARSS46834.2022.9884596. https://arxiv.org/pdf/2201.09355

12. MuraliMohanBabu Y, Subramanyam M, GiriPrasad M. A Modified BM3D Algorithm for SAR Image Despeckling. Procedia Computer Science. 2015;70: 69–75. DOIhttps://doi.org/10.1016/j.procs.2015.10.038. https://doi.org/10.1016/j.procs.2015.10.038.

13. Wang X, Ge L, Li X. Evaluation of Filters for Envisat Asar Speckle Suppression in Pasture Area. ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences. 2012;I-7: 341–346. DOIhttps://doi.org/10.5194/isprsannals-I-7-341-2012. https://isprs-annals.copernicus.org/articles/I-7/341/2012/isprsannals-I-7-341-2012.pdf

14. Sennikov AV, Larionov RV, Khryashchev VV, et al. Issledovanie neirosetevykh algoritmov dlya segmentatsii vodnykh uchastkov na sputnikovykh SAR izobrazheniyakh [Investigation of Neural Network Algorithms for Segmentation of Water Areas on Satellite SAR Images]. Digital Signal Processing. 2024;4: 83–88. (In Russian).

15. Guo M, Lu C, Hou Q, et al. SegNeXt: Rethinking Convolutional Attention Design for Semantic Segmentation. Advances in Neural Information Processing Systems 35: Proceedings of the 36th Conference on Neural Information Processing Systems (NeurIPS 2022). Red Hook: Curran Associates; 2022: 1140–1156. DOIhttps://doi.org/10.52202/068431-0084. https://www.proceedings.com/content/068/068431-0084open.pdf

16. Hanyu T, Yamazaki K, Tran M, et al. AerialFormer: Multi-Resolution Transformer for Aerial Image Segmentation. Remote Sensing. 2024; 16(16): 2930. DOI:https://doi.org/10.3390/rs16162930.

17. Setiawan AW Image Segmentation Metrics in Skin Lesion: Accuracy, Sensitivity, Specificity, Dice Coefficient, Jaccard Index, and Matthews Correlation Coefficient. 2020 International Conference on Computer Engineering, Network, and Intelligent Multimedia (CENIM). Surabaya: IEEE; 2020: 97–102. DOI:https://doi.org/10.1109/CENIM51130.2020.9297970. https://doi.org/10.1109/CENIM51130.2020.9297970.

18. Larionov R, Kotov N, Priorov A, et al. Agricultural Fields Segmentation on Satellite Images Using Convolutional Neural Networks. 2021 IEEE East-West Design and Test Symposium. Batumi: IEEE; 2021: 1–4. DOIhttps://doi.org/10.1109/EWDTS52692.2021.9581006. https://doi.org/10.1109/EWDTS52692.2021.9581006

19. Bekhtin YuS, Trinh VT. Sravnenie ehffektivnosti razlichnykh tipov veivletov pri shumopodavlenii izobrazhenii, poluchennykh s pomoshch'yu radarov s sintezirovannoi aperturoi [Comparison of the Efficiency of Different Types of Wavelets in Noise Reduction of Images Obtained with Synthetic Aperture Radars]. Aktual'nye problemy sovremennoi nauki i proizvodstva: Proceedings of the IX All-Russian Scientific and Technical Conference. Ryazan: RSREU; 2024: 342–349. (In Russian).

20. Wang J, Sun K, Cheng T, et al. Deep High-Resolution Representation Learning for Visual Recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence. 2021;43(10): 3349–3364. DOIhttps://doi.org/10.1109/TPAMI.2020.2983686. https://doi.org/10.1109/TPAMI.2020.2983686

JATS XML

Editor-in-Chief, editorial board chair

downloads

GOST

Khryaschev V. V., Larionov R. V., Priorov A. L. Hybrid Approach to Agricultural Land Segmentation on Radar Images Integrating Despeckling and Deep Learning // Izvestia Vuzov. Geodesy and Aerophotosurveying. 2026. no. 3. 3. DOI: https://doi.org/10.30533/GiA-2026-025. (Date of access 10.11.2026).

MLA

Vladimir Khryaschev, Larionov R. V., Priorov A. L.. "Hybrid Approach to Agricultural Land Segmentation on Radar Images Integrating Despeckling and Deep Learning." Izvestia Vuzov. Geodesy and Aerophotosurveying 70.3 (2026): 3. DOI: https://doi.org/10.30533/GiA-2026-025.

APA

Khryaschev V. V., Larionov R. V., Priorov A. L. (2026). Hybrid Approach to Agricultural Land Segmentation on Radar Images Integrating Despeckling and Deep Learning. Izvestia Vuzov. Geodesy and Aerophotosurveying, 70(3), 3. https://doi.org/10.30533/GiA-2026-025

Chicago

Vladimir Khryaschev, Larionov R. V., Priorov A. L.. "Hybrid Approach to Agricultural Land Segmentation on Radar Images Integrating Despeckling and Deep Learning." Izvestia Vuzov. Geodesy and Aerophotosurveying 70, no. 3 (2026): 3. https://doi.org/10.30533/GiA-2026-025.

WIKI

{{cite journal | author = Khryaschev V. V., Larionov R. V., Priorov A. L. | title = Hybrid Approach to Agricultural Land Segmentation on Radar Images Integrating Despeckling and Deep Learning | journal = Izvestia Vuzov. Geodesy and Aerophotosurveying | year = 2026 | volume = 70 | issue = 3 | pages = 3 | doi = 10.30533/GiA-2026-025 }}

Copy

Export: BibTeX EndNote RefMan RusMARC

HYBRID APPROACH TO AGRICULTURAL LAND SEGMENTATION ON RADAR IMAGES INTEGRATING DESPECKLING AND DEEP LEARNING Hybrid Approach to Agricultural Land Segmentation on Radar Images Integrating Despeckling and Deep Learning

Confirmation