Abstract:Deep neural networks (DNNs) have become essential for aerial detection. However, DNNs are vulnerable to adversarial examples, which pose great security concerns for security-critical systems. Researchers recently devised adversarial patches to evaluate the vulnerability of DNN-based aerial detection methods physically. Nonetheless, adversarial patches generated by the existing algorithms are not strong enough and extremely time-consuming. Moreover, complicated physical factors are not accommodated well during the optimization process. In this article, a novel adaptive-patch-based physical attack (AP-PA) framework is proposed to alleviate the above problems, which achieves state-of-the-art performance in both accuracy and efficiency. Specifically, AP-PA aims to generate adversarial patches that are adaptive in both physical dynamics and varying scales, and by which the particular targets can be hidden from being detected. Furthermore, the adversarial patch is also gifted with attack effectiveness against all targets of the same class with a patch outside the target (no need to smear targeted objects) and robust enough in the physical world. In addition, a new loss is devised to consider more available information of detected objects to optimize the adversarial patch, which can significantly improve the patch's attack efficacy (average precision drop up to 87.86% and 85.48% in white-box and black-box settings, respectively) and optimizing efficiency. We also establish one of the first comprehensive, coherent, and rigorous benchmarks to evaluate the attack efficacy of adversarial patches on aerial detection tasks. Finally, several proportionally scaled experiments are performed physically to demonstrate that the elaborated adversarial patches can successfully deceive aerial detection algorithms in dynamic physical circumstances. The code is available at https://github.com/JiaweiLian/AP-PA.

DorPatch: Distributed and Occlusion-Robust Adversarial Patch to Evade Certifiable Defenses

DePatch: Towards Robust Adversarial Patch for Evading Person Detectors in the Real World

CAPatch: Physical Adversarial Patch against Image Captioning Systems

PAD: Patch-Agnostic Defense against Adversarial Patch Attacks

PatchGuard++: Efficient Provable Attack Detection against Adversarial Patches

PatchGuard: A Provably Robust Defense against Adversarial Patches via Small Receptive Fields and Masking

DiffPAD: Denoising Diffusion-based Adversarial Patch Decontamination

DPatch: An Adversarial Patch Attack on Object Detectors

PatchZero: Defending against Adversarial Patch Attacks by Detecting and Zeroing the Patch

DIFFender: Diffusion-Based Adversarial Defense against Patch Attacks

DAP: A Dynamic Adversarial Patch for Evading Person Detectors

You Cannot Easily Catch Me: A Low-Detectable Adversarial Patch for Object Detectors

PatchCleanser: Certifiably Robust Defense against Adversarial Patches for Any Image Classifier

PatchCURE: Improving Certifiable Robustness, Model Utility, and Computation Efficiency of Adversarial Patch Defenses

Benchmarking Adversarial Patch Against Aerial Detection

Defensive Patches for Robust Recognition in the Physical World

Certified Defense Against Patch Attacks Via Mask-Guided Randomized Smoothing

Patch Attack for Automatic Check-out

DiffPatch: Generating Customizable Adversarial Patches using Diffusion Model

Minority Reports Defense: Defending Against Adversarial Patches

Analysis and Countermeasure Design on Adversarial Patch Attacks