El Robust Reliability for UAV Detection in Edge AI: A Data-Centric Framework via Structural–Semantic Curation, Maximin Selection, and Raspberry Pi Validation
Keywords:
. Data-Centric AI (DCAI), UAV-based Human Detection, Dataset Refinement, Semantic Data Cleaning, Design of Experiments (DoE), Perceptual Hashing (pHash), Edge AI (Raspberry Pi, NCNN).Abstract
UAV-based detection systems operating under real-world conditions exhibit failures under severe domain shifts and scale degradation, where average metrics may conceal critical vulnerabilities. This work hypothesizes that data-centric optimization, through structural deduplication (pHash/DCT) and teacher–student semantic auditing, improves worst-case robustness and cross-domain generalization in lightweight detectors without increasing model complexity under Edge AI constraints. A Data-Centric AI (DCAI) framework tailored for Edge AI is proposed, integrating: (i) perceptual deduplication via pHash to reduce structural redundancy; (ii) a teacher–student auditing scheme using YOLO11x to control semantic noise and geometric consistency; and (iii) a search space formulated as a hierarchical Design of Experiments (DoE) ablation framework (R0–R4), with minimum-recall gating and multicriteria decision-making (MCDM) selection under the Wald maximin criterion. Experimental results show limited improvements in average metrics (~1%), but consistent gains under failure scenarios, reaching up to +5.95% in precision and +1.36% in recall in the worst-case setting. GLM/ANOVA analysis demonstrates that dataset and scenario dominate performance variance (η²ₚ ≈ 0.96–0.99), while seed effects are not statistically significant (p > 0.05), confirming stability and reproducibility. Validation on Raspberry Pi 5 using NCNN demonstrates edge deployment feasibility, maintaining precision > 0.94 and recall > 0.90 at distances between 5–25 m under controlled degradation. The contribution lies in the systemic integration of a DCAI model within a hierarchical DoE framework evaluated under a maximin criterion (MCDM), enabling causal attribution of performance improvements to data design rather than to architecture under real-world Edge AI constraints for UAV applications.References
Akshatha, K. R., Karunakar, A. K., Shenoy, S. B., Phani Pavan, K., Dhareshwar, C. V., & Johnson, D. G. (2023). Manipal-UAV person detection dataset: A step towards benchmarking dataset and algorithms for small object detection. ISPRS Journal of Photogrammetry and Remote Sensing, 195, 77–89. DOI: 10.1016/j.isprsjprs.2022.11.008. ScienceDirect
Ben-Tal, A., El Ghaoui, L., & Nemirovski, A. (2009). Robust optimization. Princeton University Press. DOI: 10.1515/9781400831050. De Gruyter Brill
Bhusnoor, M., Patel, J., Mehta, A., Sainkar, S., Patel, D., & Mehendale, N. (2023). Investigating the impact of distance on object detection accuracy in unmanned aerial vehicle systems using MobileNetV3. TechRxiv. DOI: 10.36227/techrxiv.24155598.v1. d197for5662m48.cloudfront.net
Bochkovskiy, A., Wang, C.-Y., & Liao, H.-Y. M. (2020). YOLOv4: Optimal speed and accuracy of object detection. arXiv. DOI: 10.48550/arXiv.2004.10934. arXiv
Cheng, G., Ding, Q., Cai, B., Dang, C., Wang, Y., Zuo, X., & Shao, Z. (2024). CSSDet: Small object detection via cross-scale feature enhancement on drone-view images. International Journal of Digital Earth, 17(1), 2414848. DOI: 10.1080/17538947.2024.2414848. Taylor & Francis Online
Feurer, M., Eggensperger, K., Falkner, S., Lindauer, M., & Hutter, F. (2022). Auto-Sklearn 2.0: Hands-free AutoML via meta-learning. Journal of Machine Learning Research, 23(261), 1–61. jmlr.org
Fontana, R., Molena, A., Pegoraro, L., & Salmaso, L. (2023). Design of experiments and machine learning with application to industrial experiments. Statistical Papers, 64(4), 1251–1274. DOI: 10.1007/s00362-023-01437-w. Springer Nature Link
Gulrajani, I., & Lopez-Paz, D. (2021). In search of lost domain generalization. International Conference on Learning Representations. ML Anthology
Hao, Y., Pei, H., Lyu, Y., Yuan, Z., Rizzo, J.-R., Wang, Y., & Fang, Y. (2023). Understanding the impact of image quality and distance of objects to object detection performance. In 2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) (pp. 11436–11442). IEEE. DOI: 10.1109/IROS55552.2023.10342139. IEEE Xplore
Hashimoto, T. B., Srivastava, M., Namkoong, H., & Liang, P. (2018). Fairness without demographics in repeated loss minimization. In Proceedings of the 35th International Conference on Machine Learning (Vol. 80, pp. 1929–1938). PMLR. Proceedings of Machine Learning Research
Hegiste, V., Walunj, S., Antony, J., Legler, T., & Ruskowski, M. (2024). Enhancing object detection with hybrid dataset in manufacturing environments: Comparing federated learning to conventional techniques. arXiv. DOI: 10.48550/arXiv.2408.08974. arXiv
Jegham, N., Koh, C. Y., Abdelatti, M., & Hendawi, A. (2024). Evaluating the evolution of YOLO (You Only Look Once) models: A comprehensive benchmark study of YOLO11 and its predecessors. arXiv. DOI: 10.48550/arXiv.2411.00201. arXiv
Karczmarczyk, A., Wątróbski, J., Bączkiewicz, A., Mróz-Malik, O., & Drożdż, W. (2025). New robust multi-criteria decision-making method for wind farm location problems. Applied Energy, 398, 126401. DOI: 10.1016/j.apenergy.2025.126401. DOI
Khanam, R., & Hussain, M. (2024). YOLOv11: An overview of the key architectural enhancements. arXiv. DOI: 10.48550/arXiv.2410.17725. DOI
Lai, K.-T., Chung, Y.-T., Su, J.-J., Lai, C.-H., & Huang, Y.-H. (2023). AI Wings: An AIoT drone system for commanding ArduPilot UAVs. IEEE Systems Journal, 17(2), 2213–2224. DOI: 10.1109/JSYST.2022.3189011. OpenReview
Li, L., Lu, J., Zhang, S., Mohaisen, L., & Emam, M. (2023). Frame duplication forgery detection in surveillance video sequences using textural features. Electronics, 12(22), 4597. DOI: 10.3390/electronics12224597. MDPI
Lin, T.-Y., Maire, M., Belongie, S., Hays, J., Perona, P., Ramanan, D., Dollár, P., & Zitnick, C. L. (2014). Microsoft COCO: Common objects in context. In D. Fleet, T. Pajdla, B. Schiele, & T. Tuytelaars (Eds.), Computer vision – ECCV 2014 (Lecture Notes in Computer Science, Vol. 8693, pp. 740–755). Springer. DOI: 10.1007/978-3-319-10602-1_48. DOI
Northcutt, C. G., Jiang, L., & Chuang, I. L. (2021). Confident learning: Estimating uncertainty in dataset labels. Journal of Artificial Intelligence Research, 70, 1373–1411. DOI: 10.1613/JAIR.1.12125. Google Research
Oala, L., Maskey, M., Bat-Leah, L., Parrish, A., Gürel, N. M., Kuo, T.-S., Liu, Y., Dror, R., Brajovic, D., Yao, X., Bartolo, M., Gaviria Rojas, W. A., Hileman, R., Aliment, R., Mahoney, M. W., Risdal, M., Lease, M., Samek, W., Dutta, D., … Mattson, P. (2023). DMLR: Data-centric machine learning research—Past, present and future. arXiv. DOI: 10.48550/arXiv.2311.13028. Portal de Investigación Eindhoven
Rey, L., Bernardos, A. M., Dobrzycki, A. D., Carramiñana, D., Bergesio, L., Besada, J. A., & Casar, J. R. (2025). A performance analysis of You Only Look Once models for deployment on constrained computational edge devices in drone applications. Electronics, 14(3), 638. DOI: 10.3390/electronics14030638. MDPI
Sheikholeslami, S., Meister, M., Wang, T., Payberah, A. H., Vlassov, V., & Dowling, J. (2021). AutoAblation: Automated parallel ablation studies for deep learning. In The 1st Workshop on Machine Learning and Systems (EuroMLSys ’21). Association for Computing Machinery. DOI: 10.1145/3437984.3458834. DC@KTH
Taiana, M., Nascimento, J. C., & Bernardino, A. (2013). An improved labelling for the INRIA person data set for pedestrian detection. In Pattern recognition and image analysis: IbPRIA 2013 (pp. 286–295). Springer. DOI: 10.1007/978-3-642-38628-2_34. DBLP
Tan, L., Zhang, C., Bai, H., Liu, Z., & Li, Y. (2025). A real-time and efficient detector for small object in UAV aerial images. Scientific Reports, 15, 39233. DOI: 10.1038/s41598-025-22855-w. Nature
Tanaka, D., Ikami, D., Yamasaki, T., & Aizawa, K. (2018). Joint optimization framework for learning with noisy labels. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (pp. 5552–5560). IEEE. DOI: 10.1109/CVPR.2018.00582. Open Access
Triantaphyllou, E., Shu, B., Sanchez, S. N., & Ray, T. (1998). Multi-criteria decision making: An operations research approach. In J. G. Webster (Ed.), Encyclopedia of electrical and electronics engineering (Vol. 15, pp. 175–186). John Wiley & Sons. Bit CSC LSU
Zauner, C. (2010). Implementation and benchmarking of perceptual image hash functions [Master’s thesis, Upper Austria University of Applied Sciences, Hagenberg Campus]. Christoph Zauner's Blog
Zha, D., Bhat, Z. P., Lai, K.-H., Yang, F., & Hu, X. (2023). Data-centric AI: Perspectives and challenges. In Proceedings of the 2023 SIAM International Conference on Data Mining (pp. 945–948). Society for Industrial and Applied Mathematics. DOI: 10.1137/1.9781611977653.ch106. SIAM
Zhang, G., Peng, Y., & Li, J. (2025). YOLO-MARS: An enhanced YOLOv8n for small object detection in UAV aerial imagery. Sensors, 25(8), 2534. DOI: 10.3390/s25082534. PubMed
Zhu, P., Wen, L., Du, D., Bian, X., Fan, H., Hu, Q., & Ling, H. (2022). Detection and tracking meet drones challenge. IEEE Transactions on Pattern Analysis and Machine Intelligence, 44(11), 7380–7399. DOI: 10.1109/TPAMI.2021.3119563.
Downloads
Published
How to Cite
Issue
Section
License
Copyright (c) 2026 ReCIBE, electronic journal of Computing, Informatics, Biomedical and Electronics
This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.
