A Method for Enhancing the Transferability of Adversarial Examples Based on Multi-Perspective Confidence Fusion

ZHAO Changfei; DENG Xinyang; JIANG Wen; ZHU Jinbiao; GENG Jie

doi:10.12263/DZXB.20260011

您当前的位置：

首页 >

文章列表页 >

A Method for Enhancing the Transferability of Adversarial Examples Based on Multi-Perspective Confidence Fusion

更新时间：2026-05-08

- A Method for Enhancing the Transferability of Adversarial Examples Based on Multi-Perspective Confidence Fusion
- ACTA ELECTRONICA SINICA Pages: 1-16(2026)
- 作者机构：
  
  1.西北工业大学电子信息学院，陕西西安 710129
  2.中国科学院空天信息创新研究院，北京 100094
- 作者简介：
- 基金信息：
  
  National Natural Science Foundation of China(62573350);Key Research and Development Program of Shaanxi(2024QY2-GJHX-05)
- DOI：10.12263/DZXB.20260011
  CLC： TP391;
- Received：11 February 2026，
  
  Accepted：19 March 2026，
  
  Online First：07 May 2026，
- 稿件说明：
移动端阅览
ZHAO Changfei, DENG Xinyang, JIANG Wen, et al. A Method for Enhancing the Transferability of Adversarial Examples Based on Multi-Perspective Confidence Fusion[J/OL]. ACTA ELECTRONICA SINICA, 2026, 1-16.
DOI：

ZHAO Changfei, DENG Xinyang, JIANG Wen, et al. A Method for Enhancing the Transferability of Adversarial Examples Based on Multi-Perspective Confidence Fusion[J/OL]. ACTA ELECTRONICA SINICA, 2026, 1-16. DOI： 10.12263/DZXB.20260011.

摘要

对抗攻击揭示了深度学习模型的脆弱性，有效的对抗攻击方法有助于发现模型的潜在漏洞。现有的梯度对抗攻击方法过度拟合受攻击的白盒模型特性，对黑盒模型的迁移攻击性能较差。针对黑盒模型开展迁移对抗攻击研究，提出了一种基于多视角置信度融合的对抗样本迁移性提升方法，并作为通用模块嵌入到基于梯度的对抗攻击过程，以提升对抗样本的迁移性。具体而言，设计了基于双像素空间的多视角变换策略，引导模型在不同通道与空间尺度下感知图像信息，从而扩充模型的重点关注区域，针对图像形成差异化的关注分布，实现对图像信息的多视角感知；为建模视角间的冲突与不确定性，利用证据理论框架提出了基于冲突感知的置信度融合方法，从模型多个视角下的输出置信度提取预测的共性信息，避免视角特异性带来的决策干扰，有效提升模型多视角决策融合的可靠性；设计了一种双向损失优化函数，优化对抗样本偏离正确的模型决策边界，引导其处于跨视角、跨模型共享的脆弱区域，从而提升对黑盒模型的迁移攻击能力。实验表明，本文方法在跨模型架构攻击场景下能够有效提升对抗样本的迁移性，现有梯度对抗攻击组合多视角置信度融合方法后，对常规训练的卷积神经网络（Convolutional Neural Network，CNN）和Transformer架构模型的迁移攻击成功率平均提升了21.15%和13.02%，对防御模型的迁移攻击成功率平均提升了13.84%，对集成模型的迁移攻击成功率平均提升了16.14%。

Abstract

Adversarial attacks expose the vulnerabilities of deep learning models

and effective adversarial attack methods aid in uncovering potential weaknesses. Existing gradient-based adversarial attacks overfit the characteristics of attacked white-box models

resulting in poor transferability for black-box models. This paper investigates transferable adversarial attacks for black-box models

proposing a multi-perspective confidence fusion-based method to enhance the transferability of adversarial examples. This approach is integrated as a universal component into gradient-based adversarial attack processes to improve transferability. Specifically

a multi-perspective transformation strategy based on dual-pixel space is designed to guide the model in perceiving image information across different channels and spatial scales

thereby expanding the model’s areas of focus

generating a differentiated attention distribution for the image

and enabling multi-view perception of image information. To model conflicts and uncertainties across different perspectives

a conflict-aware confidence fusion method based on the evidence theory framework is proposed. This method extracts common predictive information from the confidence outputs of the model across multiple perspectives

thereby avoiding decision-making interference caused by perspective-specific biases and effectively enhancing the reliability of multi-perspective decision fusion; A bidirectional loss optimization function is designed to optimize the deviation of adversarial examples from the correct model decision boundary

guiding them to lie in the shared vulnerable regions across different views and models

thereby improving the transfer attack performance against black-box models. Experiments show that the proposed method can effectively improve the transferability of adversarial examples in cross-model architecture attack scenarios. After integrating the existing gradient-based adversarial attacks with the multi-perspective confidence fusion method

the transfer attack success rate is improved by an average of 21.15% and 13.02% for conventionally trained convolutional neural networks (CNNs) and Transformer models

respectively

by 13.84% for defense models

and by 16.14% for ensemble models.

关键词

Keywords

references

吴亚军 , 刘礼文 . 一种基于深度学习水下高速航行器的目标识别方法研究 [J ] . 指挥控制与仿真 , 2025 , 47 ( 2 ): 87 - 94 .

Wu Yajun , Liu Liwen . Research on an underwater high-speed vehicle target recognition method based on deep learning [J ] . Command Control & Simulation , 2025 , 47 ( 2 ): 87 - 94 . (in Chinese)

王浩添 , 冀振元 , 化青龙 , 等 . 基于多分支多信息多深度复值特征融合网络的SAR舰船目标识别方法 [J ] . 电子学报 , 2025 , 53 ( 10 ): 3759 - 3772 .

Wang Haotian , Ji Zhenyuan , Hua Qinglong , et al . Recognition method of ship targets for SAR based on M3Net [J ] . Acta Electronica Sinica , 2025 , 53 ( 10 ): 3759 - 3772 . (in Chinese)

Han Wenqi , Jiang Wen , Geng Jie , et al . Difference-complementary learning and label reassignment for multimodal semi-supervised Semantic segmentation of remote sensing images [J ] . IEEE Transactions on Image Processing , 2025 , 34 : 566 - 580 .

Xu Mai , Sun Xiancheng , Li Shengxi , et al . Spherical patch generative adversarial net for unconditional panoramic image generation [J ] . IEEE Transactions on Image Processing , 2025 , 34 : 3833 - 3848 .

刘文钊 , 郭凯威 . 面向深度神经网络视觉模型对抗鲁棒性的攻击与防御方法研究综述 [J ] . 网络安全技术与应用 , 2025 ( 1 ): 42 - 48 .

Liu Wenzhao , Guo Kaiwei . A review of attack and defense methods targeting the adversarial robustness of deep neural network vision models [J ] . Network Security Technology & Application , 2025 ( 1 ): 42 - 48 . (in Chinese)

刘洁怡 , 李明哲 , 杨曜铭 , 等 . 基于频域多目标优化的SAR图像对抗样本生成方法 [J ] . 电子学报 , 2025 , 53 ( 6 ): 1958 - 1968 .

Liu Jieyi , Li Mingzhe , Yang Yaoming , et al . A multi-objective optimization method in the frequency domain for SAR image adversarial sample generation [J ] . Acta Electronica Sinica , 2025 , 53 ( 6 ): 1958 - 1968 . (in Chinese)

Braiek H B , Reid T , Khomh F . Physics-guided adversarial machine learning for aircraft systems simulation [J ] . IEEE Transactions on Reliability , 2023 , 72 ( 3 ): 1161 - 1175 .

Dong Yinpeng , Liao Fangzhou , Pang Tianyu , et al . Boosting adversarial attacks with momentum [C ] // 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition . Piscataway : IEEE , 2018 : 9185 - 9193 .

Xie Cihang , Zhang Zhishuai , Zhou Yuyin , et al . Improving transferability of adversarial examples with input diversity [C ] // 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition . Piscataway : IEEE , 2019 : 2725 - 2734 .

Dong Yinpeng , Pang Tianyu , Su Hang , et al . Evading defenses to transferable adversarial examples by translation-invariant attacks [C ] // 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition . Piscataway : IEEE , 2019 : 4307 - 4316 .

冯卫栋 , 余东 , 张淳杰 , 等 . 基于扰动响应的自适应集成黑盒对抗攻击算法 [J ] . 自动化学报 , 2025 , 51 ( 8 ): 1788 - 1799 .

Feng Weidong , Yu Dong , Zhang Chunjie , et al . Perturbation response-based adaptive ensemble black-box adversarial attack algorithm [J ] . Acta Automatica Sinica , 2025 , 51 ( 8 ): 1788 - 1799 . (in Chinese)

Wang Xiaosen , Lin Jiadong , Hu Han , et al . Boosting adversarial transferability through enhanced momentum [C ] // 32nd British Machine Vision Conference . Durham : BMVA Press , 2021 .

Gao Yue , Shumailov I , Fawaz K . SEA: Shareable and explainable attribution for query-based black-box attacks [C ] // 2025 IEEE Conference on Secure and Trustworthy Machine Learning (SaTML) . Piscataway : IEEE , 2025 : 439 - 458 .

Xiong Yifeng , Lin Jiadong , Zhang Min , et al . Stochastic variance reduced ensemble adversarial attack for boosting the adversarial transferability [C ] // 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition . Piscataway : IEEE , 2022 : 14963 - 14972 .

Ma Wenshuo , Li Yidong , Jia Xiaofeng , et al . Transferable adversarial attack for both vision transformers and convolutional networks via momentum integrated gradients [C ] // 2023 IEEE/CVF International Conference on Computer Vision . Piscataway : IEEE , 2023 : 4607 - 4616 .

Liu Yanpei , Chen Xinyun , Liu Chang , et al . Delving into transferable adversarial examples and black-box attacks [C ] // 5th International Conference on Learning Representations . Toulon : ICLR , 2017 : 2235 - 2248 .

Szegedy C , Zaremba W , Sutskever I , et al . Intriguing properties of neural networks [C ] // 2nd International Conference on Learning Representations . Banff : ICLR , 2014 : 1 - 10 .

Goodfellow I J , Shlens J , Szegedy C . Explaining and harnessing adversarial examples [C ] // 3rd International Conference on Learning Representations . San Diego : ICLR , 2015 : 1 - 11 .

Kurakin A , Goodfellow I J , Bengio S . Adversarial examples in the physical world [M ] //Yampolskiy R V. Artificial intelligence safety and security . New York : Chapman and Hall/CRC , 2018 : 99 - 112 .

Li Qizhang , Guo Yiwen , Zuo Wangmeng , et al . Making substitute models more Bayesian can enhance transferability of adversarial examples [C ] // 11th International Conference on Learning Representations . Kigali : ICLR , 2023 : 37295 - 37310 .

Chen Bin , Yin Jiali , Chen Shukai , et al . An adaptive model ensemble adversarial attack for boosting adversarial transferability [C ] // 2023 IEEE/CVF International Conference on Computer Vision . Piscataway : IEEE , 2023 : 4466 - 4475 .

Gan Fuquan , Yan Wo . Boosting the transferability of adversarial examples through gradient aggregation [J ] . IEEE Transactions on Information Forensics and Security , 2025 , 20 : 5563 - 5576 .

Wang Jiafeng , Chen Zhaoyu , Jiang Kaixun , et al . Boosting the transferability of adversarial attacks with global momentum initialization [J ] . Expert Systems with Applications , 2024 , 255 : 124757 .

Li Zhankai , Wang Weiping , Li Jie , et al . Foolmix: Strengthen the transferability of adversarial examples by dual-blending and direction update strategy [J ] . IEEE Transactions on Information Forensics and Security , 2024 , 19 : 5286 - 5300 .

Wang Kunyu , He Xuanran , Wang Wenxuan , et al . Boosting adversarial transferability by block shuffle and rotation [C ] // 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition . Piscataway : IEEE , 2024 : 24336 - 24346 .

Wang Xiaosen , He Xuanran , Wang Jingdong , et al . Admix: Enhancing the transferability of adversarial attacks [C ] // 2021 IEEE/CVF International Conference on Computer Vision . Piscataway : IEEE , 2021 : 16138 - 16147 .

Qian Yaguan , Chen Kecheng , Wang Bin , et al . Enhancing transferability of adversarial examples through mixed-frequency inputs [J ] . IEEE Transactions on Information Forensics and Security , 2024 , 19 : 7633 - 7645 .

Guo Yu , Liu Weiquan , Xu Qingshan , et al . Boosting adversarial transferability through augmentation in hypothesis space [C ] // 2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition . Piscataway : IEEE , 2025 : 19175 - 19185 .

Ma Chen , Chen Li , Yong Junhai . Simulating unknown target models for query-efficient black-box attacks [C ] // 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition . Piscataway : IEEE , 2021 : 11830 - 11839 .

Yin Fei , Zhang Yong , Wu Baoyuan , et al . Generalizable black-box adversarial attack with meta learning [J ] . IEEE Transactions on Pattern Analysis and Machine Intelligence , 2024 , 46 ( 3 ): 1804 - 1818 .

郑德生 , 郑舜天 , 李晓瑜 , 等 . CBA: 基于圆几何性质的黑盒攻击方法 [J/OL ] . 计算机工程 , 1 - 18 ( 2025-03-26 )[ 2026-03-23 ] . https://doi.org/10.19678/j.issn.1000-3428.0070476 https://doi.org/10.19678/j.issn.1000-3428.0070476 .

Zheng Desheng , Zheng Shuntian , Li Xiaoyu , et al . CBA: Black box attack based on circular geometric properties [J/OL ] . Computer Engineering , 1 - 18 ( 2025-03-26 )[ 2026-03-23 ] . https://doi.org/10.19678/j.issn.1000-3428.0070476 https://doi.org/10.19678/j.issn.1000-3428.0070476 . (in Chinese)

Van de Weijer J , Gevers T , Gijsenij A . Edge-based color constancy [J ] . IEEE Transactions on Image Processing , 2007 , 16 ( 9 ): 2207 - 2214 .

Zeiler M D , Fergus R . Visualizing and understanding convolutional networks [C ] // 13th European Conference on Computer Vision . Heidelberg : Springer , 2014 : 818 - 833 .

LeCun Y , Bengio Y , Hinton G . Deep learning [J ] . Nature , 2015 , 521 ( 7553 ): 436 - 444 .

Hu Tingyu , Yin Haibing , Wang Hongkui , et al . Pixel-domain just noticeable difference modeling with heterogeneous color features [J ] . Sensors , 2023 , 23 ( 4 ): 1788 .

Russakovsky O , Deng Jia , Su Hao , et al . ImageNet large scale visual recognition challenge [J ] . International Journal of Computer Vision , 2015 , 115 ( 3 ): 211 - 252 .

Szegedy C , Vanhoucke V , Ioffe S , et al . Rethinking the inception architecture for computer vision [C ] // 2016 IEEE Conference on Computer Vision and Pattern Recognition . Piscataway : IEEE , 2016 : 2818 - 2826 .

Szegedy C , Ioffe S , Vanhoucke V , et al . Inception-v4, inception-ResNet and the impact of residual connections on learning [J ] . Proceedings of the AAAI Conference on Artificial Intelligence , 2017 , 31 ( 1 ): 4278 - 4284 .

He Kaiming , Zhang Xiangyu , Ren Shaoqing , et al . Deep residual learning for image recognition [C ] // 2016 IEEE Conference on Computer Vision and Pattern Recognition . Piscataway : IEEE , 2016 : 770 - 778 .

Tramèr F , Kurakin A , Papernot N , et al . Ensemble adversarial training: Attacks and defenses [C ] // 6th International Conference on Learning Representations . Vancouver : ICLR , 2018 : 1894 - 1913 .

Liu Zihao , Liu Qi , Liu Tao , et al . Feature distillation: DNN-oriented JPEG compression against adversarial examples [C ] // 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition . Piscataway : IEEE , 2019 : 860 - 868 .

Xu Weilin , Evans D , Qi Yanjun . Feature squeezing: Detecting adversarial examples in deep neural networks [C ] // 25th Annual Network and Distributed System Security Symposium . Rosten : The Internet Society , 2018 .

Guo Chuan , Rana M , Cissé M , et al . Countering adversarial images using input transformations [C ] // 6th International Conference on Learning Representations . Vancouver : ICLR , 2018 : 4914 - 4925 .

Xie Cihang , Wang Jianyu , Zhang Zhishuai , et al . Mitigating adversarial effects through randomization [C ] // 6th International Conference on Learning Representations . Vancouver : ICLR , 2018 : 960 - 975 .

Dosovitskiy A , Beyer L , Kolesnikov A , et al . An image is worth 16x16 words: Transformers for image recognition at scale [C ] // 9th International Conference on Learning Representations . Vienna : ICLR , 2021 : 611 - 631 .

Liu Ze , Lin Yutong , Cao Yue , et al . Swin transformer: Hierarchical vision transformer using shifted windows [C ] // 2021 IEEE/CVF International Conference on Computer Vision . Piscataway : IEEE , 2021 : 9992 - 10002 .

Liu Ze , Hu Han , Lin Yutong , et al . Swin transformer V2: Scaling up capacity and resolution [C ] // 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition . Piscataway : IEEE , 2022 : 11999 - 12009 .

Tu Zhengzhong , Talebi H , Zhang Han , et al . MaxViT: Multi-axis vision transformer [C ] // 17th European Conference on Computer Vision . Heidelberg : Springer , 2022 : 459 - 479 .

Chen Pei , Feng Zhiyong , Xing Meng , et al . Exploring imperceptible adversarial examples in YC b C r color space [C ] // 30th International Conference on Multimedia Modeling . Heidelberg : Springer , 2024 : 242 - 256 .

Views

下载量

CSCD

Alert me when the article has been cited

提交

Tools

Publicity Resources

Robust Physical Adversarial Camouflages for Image Classifiers

A Guided Diffusion-based Approach to Natural Adversarial Patch Generation

Research of Lightweight Convolution Neural Network Based on Feature Expansion Convolution

Noise-Suppression Deep Learning Strategy via Reliability Estimation Gate for Image Classification

Related Author

DUAN Ye-xin

HE Zheng-yun

ZHANG Song

ZHAN Da-zhi

WANG Tian-feng

LIN Geng-you

ZHANG Jin

PAN Zhi-song

Related Institution

Zhenjiang Campus， Army Military Transportation University

Railway Transportation College， Hunan University of Technology

Department of Cyberspace Security， Beijing Electronic Science and Technology Institute

Command and Control Engineering College， Army Engineering University

School of Cyber Science and Engineering， Wuhan University

⁰