Serbest Süzülen Füzelerin Tespitine Yönelik Veri Kümesi Üretimi ve YOLO Temelli Model Optimizasyonunun Geliştirilmesi

Öz

Serbest süzülen füzeler, düşük termal iz ve azaltılmış radar kesit alanı özellikleriyle geleneksel aktif ve pasif algılama sistemlerinin duyarlılığını sınırlandırmakta; bu durum görünür spektrumda yapay zekâ tabanlı nesne tespitini tamamlayıcı bir algılama kanalı olarak öne çıkarmaktadır. Modern harp ortamında tespit modelinin gömülü platformlarda gerçek zamanlı çalışabilmesi operasyonel etkinlik açısından kritik önem taşımakta; farklı hava koşulları, değişken bakış açıları, çoklu mühimmat senaryoları ve küçük piksel alanında hedef temsili gibi etkenler tespit performansını doğrudan etkilemektedir. Literatürde nesne tespit modellerinin başarımının görüntü düzlemindeki ölçek rejimiyle doğrudan ilişkili olduğu ve veri kümesi ile model optimizasyonunun küçük nesne rejiminde bütünleşik ele alınması gerektiği vurgulanmakla birlikte; serbest süzülen füzelere özgü kamuya açık bir veri kümesine ve bu tür veri kümelerinin üretimini küçük nesne odaklı model optimizasyonuyla birlikte tanımlayan yöntemsel bir çerçeveye rastlanılmamıştır. Bu çalışmada; tasarlanacak tespit sistemleri için serbest süzülen füzelerin görüntü düzlemindeki ölçek analizinden başlayarak küçük nesne tespiti bağlamında veri kümesi tasarımı, model optimizasyonu ve operasyonel senaryo temelli değerlendirmeyi birlikte kapsayan bütünleşik bir yöntem geliştirilmiştir. Yapılan analizlerde serbest süzülen füzelerin terminal safhada (5–7 sn penceresinde) ancak 60× optik yakınlaştırma altında küçük nesne ölçek rejimine ulaşabildiği belirlenmiş; veri üretim parametreleri, mimari müdahaleler ve değerlendirme metrikleri bu operasyonel pencere ile tutarlı biçimde tasarlanmıştır. Sunulan çerçeve; ölçek analizi, simülasyon tabanlı veri üretimi, küçük nesne odaklı YOLO optimizasyonu ve senaryo temelli sağlamlık değerlendirmesini izlenebilir biçimde birleştirerek savunma sanayisi ve akademik araştırma ortamlarında uygulanabilir, genişletilebilir bir tasarım çerçevesi sağlamaktadır.

Anahtar Kelimeler

• Serbest süzülen füze
• küçük nesne tespiti
• simülasyon tabanlı veri kümesi
• YOLO
• model optimizasyonu
• operasyonel sağlamlık.

Destekleyen Kurum

TÜBİTAK 2211-C Yurtiçi Doktora Burs Programı (Başvuru No: 1649B032501866)

Development of Dataset Production and YOLO-Based Model Optimization for the Detection of Free-Gliding Missiles

Öz

Free-gliding missiles, with their low thermal signature and reduced radar cross-section, limit the sensitivity of conventional active and passive detection systems, bringing artificial intelligence-based object detection in the visible spectrum to the agenda as a complementary detection channel. In modern warfare environments, the ability of the detection model to run in real time on embedded platforms is critical for operational effectiveness; variable weather, viewing angles, multi-munition scenarios and small-pixel target representation directly affect detection performance. The literature emphasizes that detector performance is closely related to the scale regime of the target on the image plane and that dataset design and model optimization must be addressed jointly within the small object regime. However, no publicly available dataset specific to free-gliding missiles, nor a methodological framework that defines such dataset production together with small object-focused model optimization, has been encountered. This study develops an integrated method extending from scale analysis of free-gliding missiles to dataset design, model optimization and operational scenario-based evaluation within the small object detection context. The analyses show that free-gliding missiles reach the small object scale regime consistently only under 60× optical zoom within the 5–7 s terminal window; data production parameters, architectural interventions and evaluation metrics are designed in alignment with this operational window. The proposed framework integrates scale analysis, simulation-based data generation, small object-focused YOLO optimization and scenario-based robustness assessment in a traceable manner, providing an applicable and extensible design framework for defense industry and academic research environments.

Anahtar Kelimeler

• Free-gliding missile
• small object detection
• simulation-based dataset
• YOLO
• model optimization
• operational robustness

Kaynakça

1. KAYNAKLAR (REFERENCES)
2. [1] R. Yanushevsky, Modern Missile Guidance, vol. 4, no. 1. Boca Raton: CRC Press, Taylor & Francis Group, 2008.
3. [2] A. Mahmood and F. ur Rehman, "Optimal standoff distance of subsonic unpowered gliding vehicle," Results in Control and Optimization, vol. 12, no. January, p. 100259, 2023.
4. [3] S. Wang, "Future point estimation method for unpowered gliding targets," Journal of Electronic Research and Application, vol. 8, no. 6, pp. 162–169, 2024.
5. [4] F. A. Nasir, "Passive missile detection system for aerial platform," M.S. thesis, Capital University of Science and Technology, Islamabad, 2022.
6. [5] V. L. Harshitha, K. G. S. Narayan, and J. A. Baskaradas, "A study on passive and active detection of missiles," in 2022 URSI Regional Conference on Radio Science (USRI-RCRS), 2022, pp. 4–6.
7. [6] G. Boesch, "A guide to YOLOv8 in 2024," viso.ai, 2024. [Online]. Available: https://viso.ai/deep-learning/yolov8-guide/
8. [7] J. Terven, D. M. Córdova-Esparza, and J. A. Romero-González, "A comprehensive review of YOLO architectures in computer vision: From YOLOv1 to YOLOv8 and YOLO-NAS," Machine Learning and Knowledge Extraction, vol. 5, no. 4, pp. 1680–1716, 2023.

9. [8] K. R. Akshatha et al., "Manipal-UAV person detection dataset: A step towards benchmarking dataset and algorithms for small object detection," ISPRS Journal of Photogrammetry and Remote Sensing, vol. 195, pp. 77–89, 2023.
10. [9] Z. Xiuling, W. Huijuan, S. Yu, C. Gang, Z. Suhua, and Y. Quanbo, "Starting from the structure: A review of small object detection based on deep learning," Image and Vision Computing, vol. 146, p. 105054, 2024
11. [10] H. Zhang, W. Sun, C. Sun, R. He, and Y. Zhang, "HSP-YOLOv8: UAV aerial photography small target detection algorithm," Drones, vol. 8, no. 9, 2024.
12. [11] P. Vasanthi and L. Mohan, "Efficient YOLOv8 algorithm for extreme small-scale object detection," Digital Signal Processing: A Review Journal, vol. 154, p. 104682, 2024.
13. [12] X. Zhang, T. Lu, J. Wang, S. Fu, and F. Gao, "Small object detection by edge-aware neural network," Engineering Applications of Artificial Intelligence, vol. 138, p. 109406, 2024.
14. [13] X. Zhai, Z. Huang, T. Li, H. Liu, and S. Wang, "YOLO-Drone: An optimized YOLOv8 network for tiny UAV object detection," Electronics, vol. 12, no. 17, 2023.
15. [14] O. Sahin, "Improving the performance of YOLO-based detection algorithms for small object detection in UAV-taken images," M.S. thesis, Bilkent University, Ankara, 2023.
16. [15] R. A. Zitar, M. Al-betar, M. Ryalat, and S. Kassaymeh, "A review of UAV visual detection and tracking methods," in 9th Annual Conf. on Computational Science & Computational Intelligence (CSCI'22), 2023.
17. [16] H. Genco, "Image processing based navigation in aviation," M.S. thesis, Yıldız Technical University, İstanbul, 2023.
18. [17] Rafael, "AEROSPIKE," 2022. [Online]. Available: https://www.rafael.co.il/system/aerospike/
19. [18] Raytheon, "GBU-53/B SDB II," 2010. [Online]. Available: www.airandspaceforces.com/PDF/SiteCollectionDocuments/Reports/2010/August%202010/Day25/SDBII_factsheet_0810.pdf
20. [19] U. S. Navy, "AGM-154 Joint Standoff Weapon (JSOW)," 2017. [Online]. Available: https://www.navy.mil/Resources/Fact-Files/Display-FactFiles/Article/2166748/agm-154-joint-standoff- weapon-jsow/
21. [20] MBDA, "SMARTGLIDER," 2018. [Online]. Available: https://www.mbda-systems.com/sites/mbda/files/2024-06/2018%20SMARTGLIDER%20datasheet.pdf
22. [21] Defense Express, "Operating range of Russia's new UMPB D-30 glide bomb revealed," 2024. [Online]. Available: https://en.defence-ua.com/weapon_and_tech/operating_range_of_russias_new_umpb_d_30_glide_bomb_revealed-10621.html
23. [22] EDGE, "HALCON G-120," 2026. [Online]. Available: https://edgegroupuae.com/solutions/g-120
24. [23] Rosoboronexport, "KAB-500Kr," 2026. [Online]. Available: https://roe.ru/production/aerospace-forces/air-weapons/guided-weapons/korrektiruemye-i-upravlyaemye-aviatsionnye-bomby/kab-500kr/
25. [24] Wikipedia, "Sudarshan laser-guided bomb," 2013. [Online]. Available: https://en.wikipedia.org/wiki/Sudarshan_laser-guided_bomb
26. [25] Rosoboronexport, "FAB-500 M-62," 2022. [Online]. Available: https://roe.ru/en/production/aerospace-forces/air-weapons/unguided-weapons/air-bombs/fab-500-m-62/
27. [26] Aselsan, "ASELFLIR 600," 2025. [Online]. Available: https://www.aselsan.com/tr/savunma/urun/3209/aselflir-600
28. [27] Aselsan, "ASELFLIR 400," 2025. [Online]. Available: https://www.aselsan.com/tr/savunma/urun/3207/aselflir-400
29. [28] Aselsan, "Atmaca Elektro Optik Gözetleme Sistemi," 2025. [Online]. Available: https://stage.aselsan.emrg.me/tr/savunma/urun/1684/atmaca
30. [29] Aselsan, "Kuzgun Elektro-Optik Keşif, Gözetleme ve Hedefleme Sistemi," 2025. [Online]. Available: https://stage.aselsan.emrg.me/tr/savunma/urun/1692/kuzgun
31. [30] K. Tong and Y. Wu, "Rethinking PASCAL-VOC and MS-COCO dataset for small object detection," Journal of Visual Communication and Image Representation, vol. 93, p. 103830, 2023.
32. [31] H. Bhanbhro, Y. K. Hooi, M. N. Bin Zakaria, W. Kusakunniran, and Z. H. Amur, "MCBAN: A small object detection multi-convolutional block attention network," Computers, Materials & Continua, vol. 81, no. 2, pp. 2243–2259, 2024.
33. [32] C. Dai, T. Song, Q. Chen, H. Gong, B. Yang, and G. Song, "FSIC: Frequency-separated image compression for small object detection," Digital Signal Processing: A Review Journal, vol. 156, p. 104822, 2025.
34. [33] K. Tong and Y. Wu, "Small object detection using deep feature learning and feature fusion network," Engineering Applications of Artificial Intelligence, vol. 132, p. 107931, 2024.
35. [34] B. Bischl et al., "Hyperparameter optimization: Foundations, algorithms, best practices, and open challenges," Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery, vol. 13, no. 2, 2023.
36. [35] L. N. Smith, "A disciplined approach to neural network hyper-parameters: Part 1 — learning rate, batch size, momentum, and weight decay,", US Naval Research Laboratory Technical Report 5510-026, 2018. https://doi.org/10.48550/arXiv.1803.09820.
37. [36] Arıcı, F. N., & Orhan, H., Farklı Derin Öğrenme Algoritmalarının Gerçek Zamanlı İHA Tespitine Etkileri. Gazi Üniversitesi Fen Bilimleri Dergisi Part C: Tasarım ve Teknoloji, 12(2), 691-706, 2024. https://doi.org/10.29109/gujsc.1406837.
38. [37] G. Boesch, "YOLOv7: A powerful object detection algorithm (2024 guide)," viso.ai, 2023. [Online]. Available: https://viso.ai/deep-learning/yolov7-guide/