.png){kind=icon}
.png){kind=icon}
Return
A low-frequency adversarial attack method for object detection using generative model
.png){kind=icon}
PRE
AI
DOI：10.1007/s11042-024-18189-w.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Object detection is widely employed in security-critical scenarios. With the rapid development of deep learning, deep learning-based object detection methods have gradually replaced the traditional object detection technology due to their higher efficiency and accuracy in detection. However, these deep learning-based models are vulnerable to adversarial examples, which pose a serious security threat. Currently, existing adversarial attack methods have limited attack ability and are time-consuming. To address this issue, a low-frequency adversarial examples generation method for object detection using a generative model is proposed. By transforming the generation of adversarial examples from a traditional optimization mechanism into a generation mechanism, our method greatly shortens the time required for generating adversarial examples. Two auxiliary networks are added to the Generative Adversarial Networks framework to guide the network training, using adversarial loss and the feature layer loss to improve the attack ability of adversarial examples. Moreover, a Gaussian Filtering Module is incorporated behind the generator to smooth the perturbation and preserve effective low-frequency perturbation. Experiment results on PASCAL VOC 2007 datasets show that our method can significantly improve generation speed and attack success rate compared to other attack methods. Furthermore, compared with the UEA methods, which also use a generation mechanism, our method exhibits superior performance in terms of generated image quality and attack success rate.
Keywords:
Adversarial perturbation
Object detection
Generative adversarial network
Low frequency
Journal
IF:
3
Papers:
1.9W
Citations:
3.2W