Manipulating Infrared Emissivity with Galvanized Iron Sheets for Physical Adversarial Attack

Hao Li, Guodong Xu, Maoguo Gong, Yue Wu

Keywords: Adversarial Patch, Deep neural network, Physical sample generation

TL;DR: Infrared emissivity is considered as a new physical factor for adversarial attack and galvanized iron sheets with different degrees of roughness are used to control the emissivity of the object's surface.

Abstract: For adversarial attacks on infrared detectors, previous works have focused on designing the physical patches through temperature variations, overlooking the impact of infrared emissivity on infrared imaging. In fact, infrared emissivity significantly affects infrared radiant intensity at the same temperature. In this paper, a QR-like adversarial attack patch is designed by manipulating the surface emissivity of objects to alter the infrared radiation intensity emitted from the object's surface, called Emissivity QR-like Patch (E-QR patch). In this paper, the surface emissivity of the object is manipulated through the adjustment of surface roughness. Various levels of surface roughness are realized by a commonly used metal material, galvanized iron sheets, to produce physically adversarial patches with diverse infrared radiation intensity. Considering the possible transformation distributions between the digital and physical domains, a physical E-QR patch, which is robust to noise, angle, and position, is generated by an expectation over the transformation framework. Smoothing loss is incorporated to minimize the loss in physical reconstruction, thereby effectively mitigating shooting errors in the physical domain induced by abrupt pixel changes in the digital domain. Experimental results show that the E-QR patch achieves more than 80% attack success rate for infrared pedestrian detectors in a physical environment.

Primary Area: applications to computer vision, audio, language, and other modalities

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 9895