S.A. Nenashev1, I.A. Kirshina2, T.A. Pisklenov3, V.A. Nenashev4

1–4 St. Petersburg State University of Aerospace Instrument Engineering (St. Petersburg, Russia)

1 nenashev_sergey178@mail.ru, 2 zlata@guap.ru, 3 tim.kirp@mail.ru, 4 nenashev.va@yandex.ru

Monitoring of critical infrastructure, including power lines and security zones, requires automated processing of large volumes of three-dimensional data received from on-board lidar systems. Traditional point cloud processing methods demonstrate limited efficiency when working with large-scale datasets containing billions of points of high geometric complexity. This makes it difficult to automatically identify vegetation in protected areas and other infrastructure elements. It is necessary to transform unstructured three-dimensional data into information models with explicit separation of objects using modern neural network methods. To conduct a comparative analysis of modern neural network architectures for semantic segmentation of point clouds formed by onboard lidar systems, and to develop an effective processing methodology for automating monitoring of critical infrastructure. A comparative analysis of the neural network architectures PointNet++, RandLA-Net and KPConv for semantic segmentation on point clouds has been performed. Various approaches to the preprocessing of point clouds have been investigated, including coordinate normalization, noise filtering, balanced class weighting, and optimal downsampling. A methodology for training neural network models using the Adam optimizer and weighted cross-entropy on a large-scale DALES dataset has been developed. It was found that KPConv architecture demonstrated the best performance on all classes of objects, significantly outperforming competing solutions. RandLA-Net has shown competitive results with good computational efficiency. Optimal downsampling provided a significant acceleration of processing without reducing the quality of segmentation. The presented methodology can be applied in the creation of automated infrastructure monitoring systems, systematization of security zone control and optimization of maintenance costs for power transmission lines. The obtained results provide a scientific justification for the implementation of effective solutions in the field of automatic analysis of three-dimensional data in management and control systems of critical infrastructure.

Nenashev S.A., Kirshina I.A., Pisklenov T.A., Nenashev V.A. Investigation of semantic segmentation methods based on points formed using an on-board lidar. Electromagnetic waves and electronic systems. 2026. V. 31. № 1. P. 28−37. DOI: https://doi.org/10.18127/ j15604128-202601-03 (in Russian)

1. Kovyazin V.F., Vinogradov K.P., Kitcenko A.A., Vasilyeva E.A. Airborne Laser Scanning for Clarification of the Valuation Indicators of Forest Stands. Russian Forestry Journal. 2020. № 6. P. 42–54. DOI 10.37482/0536-1036-2020-6-42-54. (in Russian)
2. Vasanov A.E., Shlyakhova M.M. Aerial scanning method. Interexpo Geo-Siberia. 2023. V. 4. № 1. P. 162–166. DOI 10.33764/2618-981X-2023-4-1-162-166. (in Russian)
3. Ryabukhin P.B., Runova E.M. The use of unmanned aerial vehicles for taxation of forest stands. Systems. Methods. Technologies. 2025. № 1(65). P. 163–171. DOI 10.18324/2077-5415-2025-1-163-171. (in Russian)
4. Valiev V.S., Ivanov D.V. Gradient algorithm for semantic segmentation of Earth surface images and its use for water bodies boundary detection. Russian Journal of Applied Ecology. 2025. № 1(41). P. 19–26. DOI 10.24852/2411-7374.2025.1.19.26. (in Russian)
5. Al-Najjar A., Amini M., Rajan S., Green J.R. Identifying Areas of High-Risk Vegetation Encroachment on Electrical Powerlines Using Mobile and Airborne Laser Scanned Point Clouds. IEEE Sensors Journal. 2024. V. 24. № 14. P. 22129–22143. DOI 10.1109/JSEN.2023. 3348785.
6. Varney N., Asari V.K., Graehling Q. DALES: A Large-scale Aerial LiDAR Data Set for Semantic Segmentation. IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). Seattle, WA, USA. 2020. P. 717–726. DOI 10.1109/CVPRW50498. 2020.00101.
7. Lukashik D.V. Analysis of modern image segmentation methods. Economics and quality of communication systems. 2022. № 2(24). P. 57–65. (in Russian)
8. Sai S.V., Zinkevich A.V. Method of semantic segmentation of airborne laser scanning data of water protection zones. Scientific and Technical Bulletin of Information Technologies, Mechanics and Optics. 2025. V. 25. № 1. P. 68–77. DOI 10.17586/2226-1494-2025-25-1-68-77. (in Russian)
9. Alekseev P.P., Kvyatkovskaya I.Yu. Applying neural networks in recognizing conditionally graphical electrical symbols. Vestnik of Astrakhan State Technical University. Series: Management, Computer Science and Informatics. 2021. № 2. P. 47–56. DOI 10.24143/ 2072-9502-2021-2-47-56. (in Russian)
10. Starovoytov A.V., Fattakhov A.V., Yachmeneva E.A., Khamiev M.M., Kisler D.A., Kosarev V.E., Nurgaliev D.K. Felling outturn assessment using Earth remote sensing data. Scientific notes of Kazan University. Series: Natural Sciences. 2021. V. 163. № 4. P. 591–602. DOI 10.26907/2542-064X.2021.4.591-602. (in Russian)
11. Burdin A.A., Bogatyrev D.A., Tarasov A.V., Pyankov S.V. Assessment of tree parameters based on aerial laser scanning data in mixed forests of the Middle Urals. Bulletin of Siberian State University of Geosystems and Technologies. 2025. V. 30. № 4. P. 32–41. DOI 10.33764/2411-1759-2025-30-4-32-41. (in Russian)
12. Leshin O.D., Grigoriev D.S. Development of a neural network model for semantic segmentation of point clouds. Proceedings of the International Conference on Computer Graphics and Vision "Graphicon". 2022. № 32. P. 1084–1089. DOI 10.20948/graphicon-2022-1084-1089. (in Russian)
13. Golubtsov V.K., Shlyakhova M.M. Development of a neural network model for automatic object classification based on LiDAR data. Interexpo Geo-Siberia. 2025. V. 6. P. 186–190. DOI 10.33764/2618-981X-2025-6-186-190. (in Russian)
14. Li J., Yang B., Chen C., Wu W., Zhang L. Aerial-triangulation aided boresight calibration for a low-cost UAV-lidar system. ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences. 2020. V. V-1-2020. P. 245–252. DOI 10.5194/isprs-annals-v-1-2020-245-2020.
15. Vasilenko D.V. Development of an algorithm for urban scale point cloud classification. Bulletin of Siberian State University of Geosystems and Technologies. 2024. V. 29. № 6. P. 44–52. DOI 10.33764/2411-1759-2024-29-6-44-52. (in Russian)
16. Mоchalov S.A. Methodological approaches to optimization the level of autonomy of radio-electronic equipment of a complex with UAV and the conditions of their application. Radiotekhnika. 2025. V. 89. № 8. P. 149−159. DOI 10.18127/j00338486-202508-17 (in Russian)
17. Makovetskiy A.Yu., Kober V.I., Voronin S.M., Voronin A.V., Karnaukhov V.N., Mozerov M.G. Point cloud registration in 3D space using soft matching. Information processes. 2024. V. 24. № 1. P. 105–117. DOI 10.53921/18195822_2024_24_1_105. (in Russian)
18. Nenashev S.A., Pisklenov T.A., Zalishchuk A.A., Nenashev V.A. Development and research of a hardware and software system based on a single-board computer using neural network models for the task of recognizing objects of interest on the Earth's surface. Radiotekhnika. 2025. V. 89. № 8. P. 15−27. DOI 10.18127/j00338486-202508-02. (in Russian)
19. Sentsov A.A., Nenashev V.A., Ivanov S.A., Turnetskaya E.L. Combining the generated radar images with a digital map of the area in on-board systems for operational monitoring of the earth's surface. Proceedings of MAI. 2021. № 117. P. 8. DOI 10.34759/trd-2021-117-08. (in Russian)
20. Nenashev V.A., Sentsov A.A. Spatially distributed radar and optical monitoring systems. Saint Petersburg: Saint Petersburg State University of Aerospace Instrumentation. 2022. 191 p. (in Russian)