Facebook pixel tracking

The Science and Information (SAI) Organization publishes open-access peer-reviewed journals in computer science and artificial intelligence.

Contact Info
Website thesai.org
Follow Us
Contact Info
Follow Us
Research Article | Open Access |

FD-Square-Det: Frequency-Guided Robust Pedestrian Detection Under Adversarial and Cross-Domain Conditions

Author 1: Prithiraj Modak Author 2: Monjurul Islam Nimon Author 3: Tanvir Ahmed Author 4: Rifath Mahmud Author 5: Syeda Anika Tasnim
International Journal of Advanced Computer Science and Applications (IJACSA) · Vol. 17, No. 10 · Published 2026

DOI: 10.14569/IJACSA.2026.0171088

Abstract

Pedestrian detectors face physical adversarial patches and distribution shift between training and deployment data. These threats are usually studied separately. This study asks whether robustness to patches and cross-domain generalization can improve together or conflict. Detectors based on You Only Look Once version 8 medium (YOLOv8m) are evaluated under a four-quadrant protocol with clean and patched images from PASCAL Visual Object Classes (VOC) and from WiderPerson, an unlabeled target domain. Scores are average precision at an intersection over union threshold of 0.5 (AP50) in the style of Common Objects in Context (COCO), with three seeds, equal training budgets, and paired statistics. The Frequency-guided Domain-aware Detector (FD-square-Det) combines a learnable frequency filter, a frequency-statistic channel attention, a gradient reversal domain branch, and patch-augmented training. Each component is audited against matched baselines. Under a universal 128-pixel patch crafted on a surrogate detector, patch training alone reduces the relative AP50 drop on VOC from 14.9 percent to 1.6 percent without changing clean AP50. The domain branch lowers clean target AP50 by about 5.6 points, and the frequency modules give no gain beyond seed noise. White-box patches optimized for each detector remain effective, with attack success between 61 and 85 percent. An explainable artificial intelligence analysis with patch-aware Gradient-weighted Class Activation Mapping (Grad-CAM) shows no better pedestrian localization for FD-square-Det. The two goals therefore need not conflict, and the tested adaptation and frequency components do not link them.

Keywords

How to Cite this Article

Modak, P., Nimon, M. I., Ahmed, T., Mahmud, R., & Tasnim, S. A. (2026). FD-Square-Det: Frequency-Guided Robust Pedestrian Detection Under Adversarial and Cross-Domain Conditions. International Journal of Advanced Computer Science and Applications, 17(10). https://doi.org/10.14569/IJACSA.2026.0171088

Modak, Prithiraj, et al.. "FD-Square-Det: Frequency-Guided Robust Pedestrian Detection Under Adversarial and Cross-Domain Conditions." International Journal of Advanced Computer Science and Applications, vol. 17, no. 10, 2026, https://doi.org/10.14569/IJACSA.2026.0171088.

@article{Modak2026,
  title     = {FD-Square-Det: Frequency-Guided Robust Pedestrian Detection Under Adversarial and Cross-Domain Conditions},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {17},
  number    = {10},
  year      = {2026},
  publisher = {The Science and Information Organization},
  author    = {Prithiraj Modak and Monjurul Islam Nimon and Tanvir Ahmed and Rifath Mahmud and Syeda Anika Tasnim},
  doi       = {10.14569/IJACSA.2026.0171088},
  url       = {https://doi.org/10.14569/IJACSA.2026.0171088}
}

Open Access — licensed under a Creative Commons Attribution 4.0 International License. Unrestricted use, distribution, and reproduction in any medium, even commercially, as long as the original work is properly cited.

Call for Papers - Important Dates

Volume 17 No. 12, December 2026
Paper Submission Due November 25, 2026
Review Notification December 15, 2026
Publication Date December 31, 2026