CCA: A Novel Camouflage Coating Attack on Object Detectors in Remote Sensing Images

Adversarial patch-based physical attacks have attracted growing attention, yet their effectiveness in deceiving aerial detection systems remains underexplored. Existing methods typically rely on carefully designed adversarial patches with specific shapes, strategically placed in or around the target. However, for small-scale remote sensing objects, these approaches often suffer from limited attack efficacy and poor transferability. Furthermore, the conspicuous nature of these patches makes them easily identifiable by human observers, undermining the fundamental principle of adversarial examples. To address these challenges, this article proposes a novel camouflage coating attack (CCA), a physical attack method specifically designed for remote sensing targets, exhibiting high attack effectiveness and strong transferability. During training, the adversarial coating is precisely applied to the target using a dedicated coating application algorithm, followed by pixel-level iterative optimization to enhance its real-world attack efficacy. To improve stealthiness, we introduce a sophisticated loss function that generates covert adversarial patterns while preserving camouflage characteristics. We rigorously evaluate CCA through black-box and white-box attack experiments on multiple object detectors, along with transferability assessments and perceptibility analyses of the generated camouflage patterns. Experimental results demonstrate that CCA achieves an attack success rate (ASR) of up to 82.96%, reducing the average precision (AP) to as low as 14.12%. This study establishes a new benchmark for assessing the adversarial robustness of object detectors and their defense mechanisms. By introducing this novel attack paradigm, we aim to inspire future advancements and challenges in adversarial research in the domains of remote sensing and object detection.