Adversarial patch attacks against aerial imagery object detectors

被引:18
作者
Tang, Guijian [1 ,2 ]
Jiang, Tingsong [2 ]
Zhou, Weien [2 ]
Li, Chao [2 ,3 ]
Yao, Wen [2 ]
Zhao, Yong [1 ]
机构
[1] Natl Univ Def Technol, Coll Aerosp Sci & Engn, 109 Deya Rd, Changsha 410073, Peoples R China
[2] Chinese Acad Mil Sci, Def Innovat Inst, 53 Fengtai East St, Beijing 100071, Peoples R China
[3] Xidian Univ, Sch Artificial Intelligence, Xian 710071, Peoples R China
基金
中国国家自然科学基金;
关键词
Adversarial patch attacks; Aerial imagery; Object detection; Black; -box;
D O I
10.1016/j.neucom.2023.03.050
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Although Deep Neural Networks (DNNs)-based object detectors are widely used in various fields, espe-cially on aerial imagery object detections, it has been observed that a small elaborately designed patch attached to the images can mislead the DNNs-based detectors into producing erroneous output. However, the target detectors being attacked are quite simple, and the attack efficiency is relatively low in previous works, making it not practicable in real scenarios. To address these limitations, a new adversarial patch attack algorithm is proposed in this paper. Firstly, we designed a novel loss function using the intermediate outputs of the models rather than the model's final outputs interpreted by the detection head to optimize adversarial patches. The experiments conducted on the DOTA, RSOD, and NWPU VHR-10 datasets demonstrate that our method can significantly degrade the performance of the detectors. Secondly, we conducted intensive experiments to investigate the impact of different out-puts of the detection model on generating adversarial patches, demonstrating the class score is not as effective as the objectness score. Thirdly, we comprehensively analyzed the attack transferability across different aerial imagery datasets, verifying that the patches generated on one dataset are also effective in attacking another. Moreover, we proposed ensemble training to boost the attack's transferability across models. Our work alarms the application of DNNs-based object detectors in aerial imagery.(c) 2023 Elsevier B.V. All rights reserved.
引用
收藏
页码:128 / 140
页数:13
相关论文
共 50 条
[41]   YOLOv5-Based Object Detection for Emergency Response in Aerial Imagery [J].
Boddu, Sindhu ;
Mukherjee, Arindam ;
Agarwal, Manan .
SOUTHEASTCON 2025, 2025, :1536-1541
[42]   Adversarial defenses for object detectors based on Gabor convolutional layers [J].
Amirkhani, Abdollah ;
Karimi, Mohammad Parsa .
VISUAL COMPUTER, 2022, 38 (06) :1929-1944
[43]   Location-independent adversarial patch generation for object detection [J].
Ding, Zhiyi ;
Sun, Lei ;
Mao, Xiuqing ;
Dai, Leyu ;
Xu, Bayi .
JOURNAL OF ELECTRONIC IMAGING, 2023, 32 (04)
[44]   TACO: Adversarial Camouflage Optimization on Trucks to Fool Object Detectors [J].
Dimitriu, Adonisz ;
Michaletzky, Tamas Vilmos ;
Remeli, Viktor .
BIG DATA AND COGNITIVE COMPUTING, 2025, 9 (03)
[45]   Adversarial attacks on Faster R-CNN object detector [J].
Wang, Yutong ;
Wang, Kunfeng ;
Zhu, Zhanxing ;
Wang, Fei-Yue .
NEUROCOMPUTING, 2020, 382 :87-95
[46]   Evaluation of synthetic aerial imagery using unconditional generative adversarial networks [J].
Yates, Matthew ;
Hart, Glen ;
Houghton, Robert ;
Torres, Mercedes Torres ;
Pound, Michael .
ISPRS JOURNAL OF PHOTOGRAMMETRY AND REMOTE SENSING, 2022, 190 :231-251
[47]   Adversarial defenses for object detectors based on Gabor convolutional layers [J].
Abdollah Amirkhani ;
Mohammad Parsa Karimi .
The Visual Computer, 2022, 38 :1929-1944
[48]   DETECTSEC: Evaluating the robustness of object detection models to adversarial attacks [J].
Du, Tianyu ;
Ji, Shouling ;
Wang, Bo ;
He, Sirui ;
Li, Jinfeng ;
Li, Bo ;
Wei, Tao ;
Jia, Yunhan ;
Beyah, Raheem ;
Wang, Ting .
INTERNATIONAL JOURNAL OF INTELLIGENT SYSTEMS, 2022, 37 (09) :6463-6492
[49]   Non-targeted Adversarial Attacks on Object Detection Models [J].
Mi, Jian-Xun ;
Zhao, Xiangjin ;
Chen, Yongtao ;
Cheng, Xiao ;
Tian, Peng ;
Lv, Xiaohong ;
Zhong, Jiayong .
ADVANCED INTELLIGENT COMPUTING TECHNOLOGY AND APPLICATIONS, PT IX, ICIC 2024, 2024, 14870 :3-12
[50]   SSIM-Based Autoencoder Modeling to Defeat Adversarial Patch Attacks [J].
Lee, Seungyeol ;
Hong, Seongwoo ;
Kim, Gwangyeol ;
Ha, Jaecheol .
SENSORS, 2024, 24 (19)