Analisis Trade-Off Presisi Recall Modifikasi YOLOv8s Lapisan P2, CBAM, Wise-IoU v3 Deteksi Objek Mikro UAV

Agung Ramadhanu; Maheza  Novrayuda

doi:10.35134/komtekinfo.v10i4.720

Authors

Agung Ramadhanu Universitas Putra Indonesia YPTK Padang
Maheza Novrayuda Universitas Putra Indonesia YPTK Padang

DOI:

https://doi.org/10.35134/komtekinfo.v10i4.720

Keywords:

Small object detection, YOLOv8, CBAM, Wise-IoU, precision–recall trade-off

Abstract

Detecting micro-sized objects in unmanned aerial vehicle (UAV) images is difficult because each object covers only a few pixels. This study examines whether three modifications of YOLOv8s increase detection precision: an extra P2 detection head (stride 4), the Convolutional Block Attention Module (CBAM), and the Wise-IoU v3 loss. In the VisDrone training labels used here, 84.80% of 22,568 annotated objects are smaller than 32 × 32 pixels. Four models (M1–M4) were trained in a controlled ablation for 20 epochs with batch size 4 on an NVIDIA GeForce RTX 2050 (4 GB) and evaluated on 98 test images. The full model (M4) raised precision from 37.90% to 48.68% (+10.78 points), but recall fell from 27.74% to 19.26%, mAP@50 from 25.86% to 15.96%, and mAPsmall from 21.98% to 14.16%. A paired bootstrap (1,000 resamples) gave a per-image F1 difference of −5.56 points (95% CI −6.89 to −4.20). On one dense test image, false positives fell from 51 to 13. The training curves show that the P2 models started from a much higher loss and were still improving at epoch 20, so unequal weight initialization is a plausible but untested explanation. M4 ran at 37.1 FPS (26.99 ms) on the RTX 2050. The findings are exploratory and describe a high-precision, low-recall trade-off rather than an overall accuracy gain.

References

P. Zhu, L. Wen, D. Du, X. Bian, H. Fan, Q. Hu, and H. Ling, "Detection and tracking meet drones challenge," IEEE Trans. Pattern Anal. Mach. Intell., vol. 44, no. 11, pp. 7380–7399, 2022, doi: 10.1109/TPAMI.2021.3119563.

T.-Y. Lin et al., "Microsoft COCO: Common objects in context," in Proc. Eur. Conf. Comput. Vis. (ECCV), 2014, pp. 740–755, doi: 10.1007/978-3-319-10602-1_48.

J. Redmon, S. Divvala, R. Girshick, and A. Farhadi, "You only look once: Unified, real-time object detection," in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), 2016, pp. 779–788, doi: 10.1109/CVPR.2016.91.

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," Mach. Learn. Knowl. Extr., vol. 5, no. 4, pp. 1680–1716, 2023, doi: 10.3390/make5040083.

G. Jocher, A. Chaurasia, and J. Qiu, "Ultralytics YOLOv8," ver. 8.0.0, 2023. [Online]. Available: https://github.com/ultralytics/ultralytics.

S. Tan, Z. Duan, and L. Pu, "Multi-scale object detection in UAV images based on adaptive feature fusion," PLoS ONE, vol. 19, no. 3, art. e0300120, 2024, doi: 10.1371/journal.pone.0300120.

L. Xu, Y. Zhao, Y. Zhai, L. Huang, and C. Ruan, "Small object detection in UAV images based on YOLOv8n," Int. J. Comput. Intell. Syst., vol. 17, no. 1, 2024, doi: 10.1007/s44196-024-00632-3.

G. Wang, Y. Chen, P. An, H. Hong, J. Hu, and T. Huang, "UAV-YOLOv8: A small-object-detection model based on improved YOLOv8 for UAV aerial photography scenarios," Sensors, vol. 23, no. 16, art. 7190, 2023, doi: 10.3390/s23167190.

G. Zhang, Y. Peng, and J. Li, "YOLO-MARS: An enhanced YOLOv8n for small object detection in UAV aerial imagery," Sensors, vol. 25, no. 8, art. 2534, 2025, doi: 10.3390/s25082534.

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, art. 3664, 2023, doi: 10.3390/electronics12173664.

F. C. Akyon, S. O. Altinuc, and A. Temizel, "Slicing aided hyper inference and fine-tuning for small object detection," in Proc. IEEE Int. Conf. Image Process. (ICIP), 2022, pp. 966–970, doi: 10.1109/ICIP46576.2022.9897990.

T.-Y. Lin, P. Dollár, R. Girshick, K. He, B. Hariharan, and S. Belongie, "Feature pyramid networks for object detection," in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), 2017, pp. 936–944, doi: 10.1109/CVPR.2017.106.

S. Woo, J. Park, J.-Y. Lee, and I. S. Kweon, "CBAM: Convolutional block attention module," in Proc. Eur. Conf. Comput. Vis. (ECCV), 2018, pp. 3–19, doi: 10.1007/978-3-030-01234-2_1.

Z. Tong, Y. Chen, Z. Xu, and R. Yu, "Wise-IoU: Bounding box regression loss with dynamic focusing mechanism," arXiv preprint arXiv:2301.10051, 2023, doi: 10.48550/arXiv.2301.10051.

Z. Zheng, P. Wang, D. Ren, W. Liu, R. Ye, Q. Hu, and W. Zuo, "Enhancing geometric factors in model learning and inference for object detection and instance segmentation," IEEE Trans. Cybern., vol. 52, no. 8, pp. 8574–8586, 2022, doi: 10.1109/TCYB.2021.3095305.

Z. Zheng, P. Wang, W. Liu, J. Li, R. Ye, and D. Ren, "Distance-IoU loss: Faster and better learning for bounding box regression," in Proc. AAAI Conf. Artif. Intell., vol. 34, no. 7, 2020, pp. 12993–13000, doi: 10.1609/aaai.v34i07.6999.

I. P. Sary, S. Andromeda, and E. U. Armin, "Performance comparison of YOLOv5 and YOLOv8 architectures in human detection using aerial images," Ultima Comput. J. Sist. Komput., vol. 15, no. 1, pp. 8–13, 2023, doi: 10.31937/sk.v15i1.3204.

N. M. Muriyah, J. H. Sim, and A. Yulianto, "Evaluating YOLOv5 and YOLOv8: Advancements in human detection," J. Inf. Syst. Informatics, vol. 6, no. 4, 2024, doi: 10.51519/journalisi.v6i4.944.

N. R. Sakhdiah and K. Fahmi, "Analisis perbandingan kinerja YOLOv8 dan YOLOv11 untuk deteksi UAV militer secara realtime," J. Algoritma, Logika dan Komputasi, vol. 9, no. 2, pp. 14–24, 2026, doi: 10.30813/j-alu.v9i2.10100.

I. H. Al Amin and A. Aprilino, "Implementasi algoritma YOLO dan Tesseract OCR pada sistem deteksi plat nomor otomatis," J. Teknoinfo, vol. 16, no. 1, pp. 54–59, 2022, doi: 10.33365/jti.v16i1.1522.

E. D. Hartono and B. Hardiansyah, "Improved YOLOv8 algorithms for object detection," in Proc. ICOIACE, Univ. 17 Agustus 1945 Surabaya, Surabaya, Indonesia, Oct. 31–Nov. 1, 2024. [Online]. Available: https://conference.untag-sby.ac.id/index.php/icoiace/article/view/5311.

X. Li, W. Wang, L. Wu, S. Chen, X. Hu, J. Li, J. Tang, and J. Yang, "Generalized focal loss: Learning qualified and distributed bounding boxes for dense object detection," in Adv. Neural Inf. Process. Syst. (NeurIPS), vol. 33, 2020, pp. 21002–21012.

B. Efron and R. J. Tibshirani, An Introduction to the Bootstrap. New York, NY, USA: Chapman & Hall, 1993..