vestifmИзвестия Национальной академии наук Беларуси. Серия физико-математических наукProceedings of the National Academy of Sciences of Belarus. Physics and Mathematics Series1561-24302524-2415The Republican Unitary Enterprise Publishing House "Belaruskaya Navuka"10.29235/1561-2430-2026-62-2-164-176vestifm-906Research ArticleИНФОРМАТИКАINFORMATICSАнализ многоканальных спутниковых изображений городских территорий методами глубокого обученияAnalysis of multichannel satellite images of urban areas using deep learning methodshttps://orcid.org/0009-0002-8652-1992ИнтяковаА. А.IntyakovaA. A.

Интякова Анастасия Алексеевна – студент

пр. Независимости, 4, 220030, Минск

Anastasia A. Intyakova – Student

4, Nezavisimosti Ave., 220030, Minsk

anastainty@gmail.comhttps://orcid.org/0000-0001-9404-1206АбламейкоС. В.AblameykoS. V.

Абламейко Сергей Владимирович – академик Национальной академии наук Беларуси, доктор технических наук, профессор

ул. Сурганова, 6, 220012, Минск

пр. Независимости, 4, 220030, Минск

Sergey V. Ablameyko – Academician of the National Academy of Sciences of Belarus, Dr. Sc. (Engineering), Professor

6, Surganov Str., 220012, Minsk

4, Nezavisimosti Ave., 220030, Minsk

ablameyko@yandex.byБелорусский государственный университетБеларусьBelarusian State UniversityBelarusБелорусский государственный университет; Объединенный институт проблем информатики Национальной академии наук БеларусиБеларусьUnited Institute of Informatics Problems of the National Academy of Sciences of Belarus; Belarusian State UniversityBelarus202610072026622164176Copyright © Интякова А.А., Абламейко С.В., 20262026Интякова А.А., Абламейко С.В.Intyakova A.A., Ablameyko S.V.Данная работа распространяется под лицензией Creative Commons Attribution 4.0.This work is licensed under a Creative Commons Attribution 4.0 License.https://vestifm.belnauka.by/jour/article/view/906

Рассмотрена проблема классификации растительного покрова в плотно застроенной урбанизированной среде по изображениям дистанционного зондирования Земли сверхвысокого разрешения. Целью исследования является разработка и валидация нейросетевого алгоритма для высокоточной семантической сегментации городских зеленых зон с использованием мультимодальных данных. Предложена усовершенствованная архитектура сверточной нейронной сети U-Net, модифицированная для работы с 7-канальным входным тензором (RGB, NIR, RedEdge, DSM, NDVI). В основе подхода лежит стратегия раннего слияния (Early Fusion), объединяющая спектральные измерения с перспективными структурными характеристиками (цифровой моделью поверхности, DSM). Для преодоления дисбаланса классов применена функция Focal Loss. Предложенная модель позволила надежно разделить спектрально идентичные ярусы (трава и деревья) и исключить ложные срабатывания на зеленых антропогенных объектах. Итоговый показатель Mean IoU составил 0,725. Полнота (Recall) обнаружения лесных массивов достигла 0,93, а сложного миноритарного класса «Кустарники» – 0,77. Эксперимент на независимой тестовой территории подтвердил высокую способность модели к генерализации (F1-score 0,93). Комплексирование 7 каналов информации и применение модифицированной U-Net полностью оправдано для задач точного расчета площадей насаждений и экологического мониторинга в концепции «Умный город».

This paper examines the problem of land cover classification in densely populated urban environments using ultra-high-resolution Earth observation images. The aim of the study is to develop and validate a neural network algorithm for a high-precision semantic segmentation of urban green spaces using multimodal data. An improved U-Net convolutional neural network architecture, modified for the use with a 7-channel input tensor (RGB, NIR, RedEdge, DSM, and NDVI), is proposed. The approach is based on the Early Fusion strategy, which combines spectral measurements with promising structural characteristics (digital surface model, DSM). Focal Loss is used to overcome the class imbalance. The proposed model reliably separated spectrally identical layers (grass and trees) and eliminated false positives on green anthropogenic objects. The final Mean IoU was 0.725. The recall for detecting forested areas reached 0.93, and for the complex minority class “Shrubs” it reached 0.77. The experiment on an independent test site confirmed the model’s high generalizability (F1-score 0.93). The integration of seven data channels and the use of a modified U-Net are fully justified for the tasks of accurate calculating forest areas and environmental monitoring in the Smart City concept.

семантическая сегментациягородские зеленые зоныдистанционное зондирование Землиглубокое обучениесверточные нейронные сетиU-Netцифровая модель поверхностимультиспектральные изображенияпространственная кросс-валидацияsemantic segmentationurban green spacesEarth remote sensingdeep learningconvolutional neural networksU-Netdigital surface modelmultispectral imageryspatial cross-validationReferencesNeyns, R. Mapping of Urban Vegetation with High-Resolution Remote Sensing: A Review / R. Neyns, F. Canters // Remote Sensing. – 2022. – Vol. 14, № 4. – Art. ID 1031. https://doi.org/10.3390/rs14041031Neyns, R. Mapping of Urban Vegetation with High-Resolution Remote Sensing: A Review / R. Neyns, F. Canters // Remote Sensing. – 2022. – Vol. 14, № 4. – Art. ID 1031. https://doi.org/10.3390/rs14041031Long, J. Fully convolutional networks for semantic segmentation / J. Long, E. Shelhamer, T. Darrell // Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). – Boston, MA, USA, 2015. – P. 3431–3440. https://doi.org/10.1109/cvpr.2015.7298965Long, J. Fully convolutional networks for semantic segmentation / J. Long, E. Shelhamer, T. Darrell // Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). – Boston, MA, USA, 2015. – P. 3431–3440. https://doi.org/10.1109/cvpr.2015.7298965Semantic Segmentation Using Deep Learning with Vegetation Indices for Rice Lodging Identification in Multi-date UAV Visible Images / M.-D. Yang, H.-H. Tseng, Y.-C. Hsu, H. P. Tsai // Remote Sensing. – 2020. – Vol. 12, № 4. – Art. ID 633. https://doi.org/10.3390/rs12040633Semantic Segmentation Using Deep Learning with Vegetation Indices for Rice Lodging Identification in Multi-date UAV Visible Images / M.-D. Yang, H.-H. Tseng, Y.-C. Hsu, H. P. Tsai // Remote Sensing. – 2020. – Vol. 12, № 4. – Art. ID 633. https://doi.org/10.3390/rs12040633U-Net Ensemble for Enhanced Semantic Segmentation in Remote Sensing Imagery / I. Dimitrovski, V. Spasev, S. Loshkovska, I. Kitanovski // Remote Sensing. – 2024. – Vol. 16, № 12. – Art. ID 2077. https://doi.org/10.3390/rs16122077U-Net Ensemble for Enhanced Semantic Segmentation in Remote Sensing Imagery / I. Dimitrovski, V. Spasev, S. Loshkovska, I. Kitanovski // Remote Sensing. – 2024. – Vol. 16, № 12. – Art. ID 2077. https://doi.org/10.3390/rs16122077GreenNet: A dual-encoder network for urban green space classification using high-resolution remotely sensed images / Ke Chen, Yang Wang, Cunrui Huang [et al.] // International Journal of Applied Earth Observation and Geoinformation. – 2025. – Vol. 142. – Art. ID 104709. https://doi.org/10.1016/j.jag.2025.104709GreenNet: A dual-encoder network for urban green space classification using high-resolution remotely sensed images / Ke Chen, Yang Wang, Cunrui Huang [et al.] // International Journal of Applied Earth Observation and Geoinformation. – 2025. – Vol. 142. – Art. ID 104709. https://doi.org/10.1016/j.jag.2025.104709Khan, B. A. Semantic Segmentation of Aerial Imagery Using U-Net with Self-Attention and Separable Convolutions / B. A. Khan, J.-W. Jung // Applied Sciences. – 2024. – Vol. 14, № 9. – Art. ID 3712. https://doi.org/10.3390/app14093712Khan, B. A. Semantic Segmentation of Aerial Imagery Using U-Net with Self-Attention and Separable Convolutions / B. A. Khan, J.-W. Jung // Applied Sciences. – 2024. – Vol. 14, № 9. – Art. ID 3712. https://doi.org/10.3390/app14093712Li, R. Research on CTSA-DeepLabV3+ Urban Green Space Classification Model Based on GF-2 Images / R. Li, J. Zhao, Y. Fan // Sensors. – 2025. – Vol. 25, № 13. – Art. ID 3862. https://doi.org/10.3390/s25133862Li, R. Research on CTSA-DeepLabV3+ Urban Green Space Classification Model Based on GF-2 Images / R. Li, J. Zhao, Y. Fan // Sensors. – 2025. – Vol. 25, № 13. – Art. ID 3862. https://doi.org/10.3390/s25133862Ying Wei. Semantic Segmentation of Urban Green Space and Water Body in High-Resolution Remote Sensing Images Based on Low-Rank Adaptation / Ying Wei, Chen Yang, Tianxing Yin // 37th Chinese Control and Decision Conference (CCDC). – Xiamen, China, 2025. – P. 418–423. https://doi.org/10.1109/ccdc65474.2025.11090983Ying Wei. Semantic Segmentation of Urban Green Space and Water Body in High-Resolution Remote Sensing Images Based on Low-Rank Adaptation / Ying Wei, Chen Yang, Tianxing Yin // 37th Chinese Control and Decision Conference (CCDC). – Xiamen, China, 2025. – P. 418–423. https://doi.org/10.1109/ccdc65474.2025.11090983Ronneberger, O. U-Net: Convolutional Networks for Biomedical Image Segmentation / O. Ronneberger, P. Fischer, T. Brox // Medical Image Computing and Computer-Assisted Intervention – MICCAI 2015 / eds.: N. Navab, J. Hornegger, W. Wells, A. Frangi. – Springer, Cham, 2015. – P. 234–241. – (Lecture Notes in Computer Science; vol. 9351). – https://doi.org/10.1007/978-3-319-24574-4_28Ronneberger, O. U-Net: Convolutional Networks for Biomedical Image Segmentation / O. Ronneberger, P. Fischer, T. Brox // Medical Image Computing and Computer-Assisted Intervention – MICCAI 2015 / eds.: N. Navab, J. Hornegger, W. Wells, A. Frangi. – Springer, Cham, 2015. – P. 234–241. – (Lecture Notes in Computer Science; vol. 9351). – https://doi.org/10.1007/978-3-319-24574-4_28Lee, D. H. A Review on Recent Deep Learning-Based Semantic Segmentation for Urban Greenness Measurement / D. H. Lee, H. Y. Park, J. Lee // Sensors. – 2024. – Vol. 24, № 7. – Art. ID 2245. https://doi.org/10.3390/s24072245Lee, D. H. A Review on Recent Deep Learning-Based Semantic Segmentation for Urban Greenness Measurement / D. H. Lee, H. Y. Park, J. Lee // Sensors. – 2024. – Vol. 24, № 7. – Art. ID 2245. https://doi.org/10.3390/s24072245Zhang, D. Semantic segmentation of multi-source remote sensing image based on multi-path context module / D. Zhang, Y. Yan // Discover Applied Sciences. – 2026. – Vol. 8. – Art. ID 361. https://doi.org/10.1007/s42452-026-08338-1Zhang, D. Semantic segmentation of multi-source remote sensing image based on multi-path context module / D. Zhang, Y. Yan // Discover Applied Sciences. – 2026. – Vol. 8. – Art. ID 361. https://doi.org/10.1007/s42452-026-08338-1Focal Loss for Dense Object Detection / Tsung-Yi Lin, P. Goyal, R. Girshick [et al.] // Proceedings of the IEEE International Conference on Computer Vision (ICCV). – Venice, Italy, 2017. – P. 2980–2988. https://doi.org/10.1109/iccv.2017.324Focal Loss for Dense Object Detection / Tsung-Yi Lin, P. Goyal, R. Girshick [et al.] // Proceedings of the IEEE International Conference on Computer Vision (ICCV). – Venice, Italy, 2017. – P. 2980–2988. https://doi.org/10.1109/iccv.2017.324Kingma, D. P. Adam: A Method for Stochastic Optimization / D. P. Kingma, J. Ba // Arxiv [Preprint]. – 2014. – URL: https://arxiv.org/abs/1412.6980. https://doi.org/10.48550/arXiv.1412.6980Kingma, D. P. Adam: A Method for Stochastic Optimization / D. P. Kingma, J. Ba // Arxiv [Preprint]. – 2014. – URL: https://arxiv.org/abs/1412.6980. https://doi.org/10.48550/arXiv.1412.6980

The authors declare that there are no conflicts of interest present.