M.V. Bukin1, A.A. Kerkhaili2, A.A. Shiryaev3, A.N. Dementiev4

1,2 Military Educational and Scientific Center of the Air Force “Air Force Academy
n.a. Professor N.E. Zhukovsky and Yu.A. Gagarin” (Voronezh)
3,4 MIREA – Russian Technological University (RTU MIREA) (Moscow, Russia)
1 bukinm@mail.ru, 2 ali.ahmad.karhili@gmail.com

This investigation aims to examine the influence of radar shadows on enhancing the efficiency of detecting ground objects (especially camouflaged ones) in Synthetic Aperture Radar (SAR) images using YOLO networks. The work utilized the benchmark MSTAR dataset (8,688 images of 8 categories of military objects), split in an 80% to 20% ratio for training and testing, respectively. Two methods were applied: the first trained a YOLOv4 model with bounding boxes including the object and its shadow; the second–with the object only (without the shadow). Performance was then evaluated under ideal conditions and under the influence of various disturbances (speckle noise, electronic countermeasures, Gaussian noise) and radar camouflage.

The results showed that under ideal conditions, the inclusion of shadows reduced the False Positive (FP) rate by 43% at a detection threshold of 0.6 and increased Precision from 90.95% to 93.52%, while the balanced F1-Score improved by 1.52%. Under camouflage conditions, the model trained with shadows demonstrated a 19.73% increase in Recall and a 21.74% improvement in mAP, with a significant (77%) reduction in missed objects (False Negatives – FN) compared to the model without shadows. In contrast, the model without shadows showed superiority under the influence of disturbances (especially electronic countermeasures and Gaussian noise), where the Recall of the model with shadows decreased by 14.62% under Gaussian noise due to the distortion of subtle shadow details.

The investigation concluded that shadow is a critically important factor for detecting camouflaged objects in SAR images; however, in the presence of disturbances, it becomes a hindrance, necessitating its disablement under strong noise conditions.

Bukin M.V., Kerkhaili A.A., Shiryaev A.A., Dementiev A.N. Investigation of the influence of shadows on the detection of ground objects in synthetic aperture radar (SAR) images. Science Intensive Technologies. 2025. V. 26. № 6. P. 5−17. DOI: https://doi.org/ 10.18127/j19998465-202506-01 (in Russian)

1. Wang L., Li Y., Zhang Z., Wang H. Robust Vehicle Detection in SAR Images Based on Shadow Feature and Multi-Level Fusion. IEEE Geoscience and Remote Sensing Letters. 2022. V. 19. Art. № 6503505. https://doi.org/10.1109/LGRS.2022.3176144.
2. Moreira A., Prats-Iraola P., Younis M., Krieger G., Hajnsek I., Papathanassiou K.P. A Tutorial on Synthetic Aperture Radar. IEEE Geoscience and Remote Sensing Magazine. 2023. V. 11. № 1. P. 6–43. https://doi.org/10.1109/MGRS.2023.3236363.
3. Chen J., Li Q., Zhang Y. Spectral Discrepancy Analysis for SAR Target Detection in Complex Backgrounds. IEEE Transactions on Geoscience and Remote Sensing. 2023. V. 61. P. 1–15. Art. № 5200815. https://doi.org/10.1109/TGRS.2023.3239876.
4. Zhao R., Wang L., Niu K. Shadow-Based Detection of Camouflaged Targets in SAR Imagery Using Multi-Feature Fusion. IEEE Transactions on Geoscience and Remote Sensing. 2023. V. 61. P. 1–13. Art. № 5201013. https://doi.org/10.1109/TGRS.2023.3298765.
5. Cumming I.G., Wong F.H. Digital Processing of Synthetic Aperture Radar Data: Algorithms and Implementation. Artech House. 2023. P. 23–25.
6. Keydel E., Lee S.W. MSTAR Public Targets and Backgrounds Data Set. U.S. Army Research Laboratory. Technical Report
   № ARL-TR-8391. 1996. Available: https://www.sdms.afrl.af.mil/datasets/mstar/.
7. Çetin M., Karl W.C. (Eds.) Deep Learning for Radar Perception. Artech House. 2023. Chap. 3: “SAR Target Detection: From Classical to Deep Learning”. P. 67–92.
8. Achim A., Camry R.I. (Eds.) Deep Learning in Radar and Communications Integration. Springer. 2023. Chap. 4: “Revolutionary Advances in SAR Target Detection”. P. 89–127.
9. Kim K.J. Military Applications of Deep Learning in Radar Imaging. Artech House. 2023. Chap. 4: “YOLO Frameworks for Real-Time Defense Systems”. P. 112–154.
10. Gupta A., Li R., Nair S. Impact of Shadow Annotation on SAR Target Recognition. IEEE Transactions on Geoscience and Remote Sensing. 2020. V. 58. № 8. P. 5632–5645. https://doi.org/10.1109/TGRS.2020.2971819.
11. Thompson P., Alvarez R., Chen M. Shadow Utilization in SAR ATR: Help or Hindrance? Proceedings of IEEE Radar Conference. 2022.
    P. 1–6. https://doi.org/10.1109/RadarConf2248738.2022.9763787.
12. Chen E., Gupta R., Wang L. Adaptive Attention for SAR Target Recognition. Nature Machine Intelligence. 2023. V. 5. № 8. P. 876–889. https://doi.org/10.1038/s42256-023-00687-5.
13. Géron A. Hands-On Machine Learning with Scikit-Learn, Keras, and TensorFlow. O'Reilly Media. 2022. Chap. 8: "Evaluation Metrics for Object Detection". P. 321–358.
14. Redmon J., Farhadi A. YOLOv3: An Incremental Improvement. In: Computer Vision: Principles, Algorithms, Applications, Learning. Academic Press. 2023. Chap. 22: “Evaluation Challenges in Object Detection”. P. 823–825.
15. Szeliski R. Computer Vision: Algorithms and Applications. Springer. 2022. Chap. 6: “Object Detection and Recognition”. P. 345–348.
16. Touzi R., Boerner W.M., Lee J.-S. Polarimetric SAR Calibration: The SLICY Target Approach. IEEE Transactions on Geoscience and Remote Sensing. 2022. V. 60. P. 1–14. Art. № 5207514. https://doi.org/10.1109/TGRS.2021.3132125.
17. Lee J.-S., Pottier E. Polarimetric Radar Imaging: From Basics to Applications. CRC Press. 2023. Chap. 3: “Speckle Noise in SAR Images”. P. 78–92.
18. Adamy D.L. EW 104: Electronic Warfare Against a New Generation of Threats. Artech House. 2023. Chap. 6: “Radar Jamming Techniques and Modeling”. P. 189–223.
19. Gonzalez R.C., Woods R.E. Digital Image Processing. Pearson. 2023. Chap. 5: “Image Restoration and Noise Models”. P. 289–324.
20. Li X., Zhang Q., Wang Y. Advanced Stealth Technologies in Modern Radar Systems: Principles and Applications. IEEE Transactions on Aerospace and Electronic Systems. 2023. V. 59. № 2. P. 1125–1143.