基于数据增强的车辆鲁棒对抗纹理生成

打开文本图片集

中图分类号：TP391.41 文献标志码：A DOI：10.12305/j.issn.1001-506X.2025.06.04

Abstract：Most of the existing physical adversarial attack methods are limited to planar patches，and even theadversarial samples that can perform multi-angle atacks sufer from insufficient robustness，insufficient generalization，anda largegap between theatack efects in the digital and physical domains.A white-box vehicleadversarial texture generation method is proposed based on this：add images with different brightness and contrast in the training dataset，and add noise that simulates the real environment on the texture generated aftereach training epoch，use the Bayesian optimization algorithm to compute the weights of the diferent loss terms，and finallyaddaregularization term to reducethe overfitting of the model.Inresponse to the problem thatthe modeland the target of the existing dataset cannot becompletely overlapped，an inpainting method is proposed for repairing images to reduce the gap between the digital simulation and the real shot.Digital simulation experiments and physical world experiments show that the proposed algorithm achieves a higher attack success rate and lower precision rate compared to existing adversarial texture generation algorithms.

Keywords：adversarial attack；physical attack；texture generation；white-box attack

0 引言

深度神经网络（deepneuralnetwork，DNN）因其强大的性能而备受关注，在计算机视觉（computervision，CV）领域有着广泛的应用。（剩余19904字）