Науковий вісник НГУ

Neural network method for invariant recognition of vehicles in aerospace images

• 
• 

User Rating:  / 0

PoorBest 

Details

Category: Content №1 2026

Last Updated on 27 February 2026

Published on 30 November -0001

Hits: 2359

Authors:

V. Yu. Kashtan, orcid.org/0000-0002-0395-5895, Dnipro University of Technology, Dnipro, Ukraine, e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.

O. V. Kazymyrenko, orcid.org/0000-0001-5506-6128, Dnipro University of Technology, Dnipro, Ukraine, e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.

V. V. Hnatushenko*, orcid.org/0000-0003-3140-3788, Dnipro University of Technology, Dnipro, Ukraine, e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.

* Corresponding author e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.

повний текст / full article

Naukovyi Visnyk Natsionalnoho Hirnychoho Universytetu. 2026, (1): 130 - 137

https://doi.org/10.33271/nvngu/2026-1/130

Abstract:

Purpose. This work proposes to develop a neural network method for invariant recognition of vehicles in high spatial resolution aerospace images using a Spatial Transformer Network.

Methodology. To ensure invariance to rotation, scale, and displacement of objects, the Spatial Transformer Network (STN) and Rotated RoI Align modules are integrated, allowing objects to be classified and localised on the presented dataset. Model optimisation is achieved by minimising a multi-task loss function that considers recognition, segmentation, and control of STN transformation parameters to prevent overfitting.

Findings. The proposed architecture combines a multi-level representation of features with a decoding module for simultaneous semantic segmentation and accurate vehicle positioning. The proposed method was evaluated by comparing it with popular object detection architectures: YOLOv8, SSD, RetinaNet, Faster R-CNN, YOLOv5, and YOLOv7, on a specialized aerospace dataset. The model demonstrated the highest and most balanced performance: accuracy = 100.0 %, FP = 0, and recall = 95.5 % (107 out of 112 vehicles detected). It significantly exceeds the performance of other neural architectures, which had either a high false positive rate (SSD) or low completeness (Faster R-CNN, 26.8 %), confirming the effectiveness of the proposed architecture.

Originality. A multi-component approach to detecting vehicles in aerospace images is proposed. It combines multi-level feature representation with Backbone Network, invariant STN mechanisms, and Rotated RoI Align. This combination ensures accurate detection of objects of arbitrary scale and rotation. Additionally, semantic segmentation of contextual information (such as roads and lanes) is applied, which increases the accuracy of object localization. The proposed multi-task loss function simultaneously optimises vehicle detection, segmentation, and stabilises STN training. As part of the study, a specialised dataset was created from images taken with a SONY DSC-WX220 camera. In this dataset, vehicles were annotated using oriented bounding boxes. This approach minimises the influence of the background and ensures correct model training.

Practical value. The developed method provides accurate and invariant detection of vehicles in aerospace images, allowing for automated assessment of traffic density and traffic flow characteristics. The technique can be used in traffic management systems.

Keywords: semantic segmentation, aerospace images, invariant recognition, convolutional neural networks

References.

1. Byun, S., Shin, I.-K., Moon, J., Kang, J., & Choi, S.-I. (2021). Road traffic monitoring from UAV images using deep learning networks. Remote Sensing, 13, 4027. https://doi.org/10.3390/rs13204027

2. Liao, W., Chen, X., Yang, J. F., Roth, S., Goesele, M., Yang, M. Y., & Rosenhahn, B. (2020). LR-CNN: Local-aware Region CNN for Vehicle Detection in Aerial Imagery. ISPRS Annals of Photogrammetry, Remote Sensing and Spatial Information Sciences, V-2–2020, 381-388. https://doi.org/10.5194/isprs-annals-V-2-2020-381-2020

3. Preethi Latha, T., Naga Sundari, K., Cherukuri, S., & Prasad, M. V. (2019). Remote Sensing UAV/Drone technology as a tool for urban development measures in APCRDA. International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, XLII-4/W16, 525-529. https://doi.org/10.5194/isprs-archives-XLII-4-W16-525-2019

4. Ivanov, D. V., Hnatushenko, V. V., Kashtan, V. Yu., & Garkusha, I. M. (2022). Computer Modeling of Territory Flooding in the Event of an Emergency at the Seredniodniprovska Hydroelectric Power Plant. Naukovyi Visnyk Natsionalnoho Hirnychoho Universytetu, (6), 123-128. https://doi.org/10.33271/nvngu/2022-6/123

5. Tang, T., Zhou, S., Deng, Z., Zou, H., & Lei, L. (2017). Vehicle Detection in Aerial Images Based on Region Convolutional Neural Networks and Hard Negative Example Mining. Sensors, 17, 336. https://doi.org/10.3390/s17020336

6. Kashtan, V., & Hnatushenko, V. (2023). Deep Learning Technology for Automatic Burned Area Extraction Using Satellite High Spatial Resolution Images. Lecture Notes in Computational Intelligence and Decision Making. Advances in Intelligent Systems and Computing, 1246, (рp. 664-685). Springer, Cham. https://doi.org/10.1007/978-3-031-16203-9_37

7. Kashtan, V., Hnatushenko, V., & Zhir, S. (2021). Information Technology Analysis of Satellite Data for Land Irrigation Monitoring. 2021 IEEE International Conference on Information and Telecommunication Technologies and Radio Electronics (UkrMiCo), (pp. 12-15). Kyiv, Ukraine. https://doi.org/10.1109/UkrMiCo52950.2021.9716592

8. Li, B., & Fang, L. (2020). Laser Radar Application in Vehicle Detection Under Traffic Environment. Proceedings of the International Conference on Artificial Intelligence and Security, (pp. 126-134). Singapore: Springer. https://doi.org/10.1007/978-981-13-9406-5_126

9. Xu, Y., Yu, G., Wang, Y., Wu, X., & Ma, Y. (2016). A Hybrid Vehicle Detection Method Based on Viola-Jones and HOG plus SVM from UAV Images. Sensors, 16, 1325. https://doi.org/10.3390/s16081325

10.      Tong, K., Wu, Y., & Zhou, F. (2020). Recent advances in small object detection based on deep learning: A review. Image and Vision Computing, 97, 103910. https://doi.org/10.1016/j.imavis.2020.103910

11.      Hsu, S. C., Huang, C. L., & Chuang, C. H. (2018). Vehicle detection using simplified fast R-CNN. 2018 International Workshop on Advanced Image Technology (IWAIT), Chiang Mai, Thailand, 1-3. https://doi.org/10.1109/IWAIT.2018.8369652

12.      Nguyen, H. (2019). Improving Faster R-CNN framework for fast vehicle detection. Mathematical Problems in Engineering, 1-11. https://doi.org/10.1155/2019/2312975

13.      Hakim, L., Hendrawan, A., & Khoiriyah, R. (2024). Traffic Vehicle Detection Using Faster R-CNN: A Comparative Analysis of Backbone Architectures. International Journal of Artificial Intelligence and Science, 1, 50-62. https://doi.org/10.63158/IJAIS.v1.i1.5

14.      Kou, J., Zhan, T., Zhou, D., Xie, Y., Da, Z., & Gong, M. (2023). Visual Attention-Based Siamese CNN with Softmaxfocal Loss for Laser-Induced Damage Change Detection of Optical Elements. Neurocomputing, 517, 173-187. https://doi.org/10.1016/j.neucom.2022.10.074

15.      Cao, C., Wang, B., Zhang, W., Zeng, X., Yan, X., Feng, Z., Liu, Y., & Wu, Z. (2019). An improved faster R-CNN for small object detection. IEEE Access, 7, 106838-106846. https://doi.org/10.1109/ACCESS.2019.2933173

16.      Kong, X., Zhang, Y., Tu, S., Xu, C., & Yang, W. (2023). Vehicle Detection in High-Resolution Aerial Images with Parallel RPN and Density-Assigner. Remote Sensing, 15, 1659. https://doi.org/10.3390/rs15061659

17.      Redmon, J., & Farhadi, A. (2018). YOLOv3: An incremental improvement. arXiv:1804.02767. https://doi.org/10.48550/arXiv.1804.02767

18.      Bochkovskiy, A., Wang, C. Y., & Liao, H. (2020). YOLOv4: Optimal Speed and Accuracy of Object Detection. arXiv:2004.10934. https://doi.org/10.48550/arXiv.2004.10934

19.      Yin, Q., Yang, W., Ran, M., & Wang, S. (2021). FD-SSD: An improved SSD object detection algorithm based on feature fusion and dilated convolution. Signal Processing: Image Communication, 98, 116402. https://doi.org/10.1016/j.image.2021.116402

20.      Hnatushenko, V., Kashtan, V., & Kazymyrenko, O. (2025). Information Technology for Detecting Cars on Aerial Imaging Using a Modified YOLO-OBB Architecture. MoDaST 2025: Modern Data Science Technologies – Doctoral Consortium, (pp. 293-304). June 15. Lviv, Ukraine. Retrieved from https://ceur-ws.org/Vol-4005/paper20.pdf

21.      Jing, R., Liu, S., Gong, Z., Wang, Z., Guan, H., Gautam, A., & Zhao, W. (2020). Object-Based Change Detection for Very High-Resolution Remote Sensing Images Based on a Trisiamese-LSTM. International Journal of Remote Sensing, 41(16), 6209-6231. https://doi.org/10.1080/01431161.2020.1734253

22.      Alif, M. A. R. (2024). YOLOv11 for Vehicle Detection: Advancements, Performance, and Applications in Intelligent Transportation Systems, 16. https://doi.org/10.48550/arXiv.2410.22898

23.      Zhao, M., Yan Zhong, Y., Sun, D., & Chen, Y. (2021). Accurate and efficient vehicle detection framework based on SSD algorithm. IET Image Processing, 15, 3094-3104. https://doi.org/10.1049/ipr2.12297

24.      Ammar, A., Koubaa, A., Ahmed, M., Saad, A., & Benjdira, B. (2021). Vehicle Detection from Aerial Images Using Deep Learning: A Comparative Study. Electronics, 10, 820. https://doi.org/10.3390/electronics10070820

25.      Ma, X., & Yang, Z. (2021). A new multi-scale backbone network for object detection based on asymmetric convolutions. Science Progress, 104, 1-17. https://doi.org/10.1177/00368504211011343

26.      Xie, X., Cheng, G., Wang, J., Yao, X., & Han, J. (2021). Oriented R-CNN for Object Detection. https://doi.org/10.48550/arXiv.2108.05699

• < Prev
• Next >