Self-Attention Context Network: Addressing the Threat of Adversarial Attacks for Hyperspectral Image Classification

被引:79
作者
Xu, Yonghao [1 ,2 ]
Du, Bo [3 ,4 ]
Zhang, Liangpei [1 ,2 ]
机构
[1] Wuhan Univ, State Key Lab Informat Engn Surveying Mapping & R, Wuhan 430079, Peoples R China
[2] Wuhan Univ, Inst Artificial Intelligence, Wuhan 430079, Peoples R China
[3] Wuhan Univ, Inst Artificial Intelligence, Natl Engn Res Ctr Multimedia Software, Sch Comp Sci, Wuhan 430079, Peoples R China
[4] Wuhan Univ, Hubei Key Lab Multimedia & Network Commun Engn, Wuhan 430079, Peoples R China
基金
中国国家自然科学基金;
关键词
Deep learning; Training; Hyperspectral imaging; Feature extraction; Task analysis; Perturbation methods; Predictive models; Hyperspectral image (HSI) classification; adversarial example; adversarial attack; adversarial defense; convolutional neural network (CNN); deep learning; NEURAL-NETWORKS; DEEP; CNN;
D O I
10.1109/TIP.2021.3118977
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Deep learning models have shown their great capability for the hyperspectral image (HSI) classification task in recent years. Nevertheless, their vulnerability towards adversarial attacks could not be neglected. In this study, we systematically analyze the influence of adversarial attacks on the HSI classification task for the first time. While existing research of adversarial attacks focuses on the generation of adversarial examples in the RGB domain, the experiments in this study show such adversarial examples could also exist in the hyperspectral domain. Although the difference between the generated adversarial image and the original hyperspectral data is imperceptible to the human visual system, most of the existing state-of-the-art deep learning models could be fooled by the adversarial image to make wrong predictions. To address this challenge, a novel self-attention context network (SACNet) is further proposed. We discover that the global context information contained in HSI can significantly improve the robustness of deep neural networks when confronted with adversarial attacks. Extensive experiments on three benchmark HSI datasets demonstrate that the proposed SACNet possesses stronger resistibility towards adversarial examples compared with the existing state-of-the-art deep learning models.
引用
收藏
页码:8671 / 8685
页数:15
相关论文
共 56 条
[1]   Threat of Adversarial Attacks on Deep Learning in Computer Vision: A Survey [J].
Akhtar, Naveed ;
Mian, Ajmal .
IEEE ACCESS, 2018, 6 :14410-14430
[2]  
[Anonymous], 2017, P ACM WORKSH ART INT, DOI DOI 10.1145/3128572.3140449
[3]   On the Robustness of Semantic Segmentation Models to Adversarial Attacks [J].
Arnab, Anurag ;
Miksik, Ondrej ;
Torr, Philip H. S. .
2018 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR), 2018, :888-897
[4]   3-D Deep Learning Approach for Remote Sensing Image Classification [J].
Ben Hamida, Amina ;
Benoit, Alexandre ;
Lambert, Patrick ;
Ben Amar, Chokri .
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, 2018, 56 (08) :4420-4434
[5]   Spectral-Spatial Classification of Hyperspectral Data Based on a Stochastic Minimum Spanning Forest Approach [J].
Bernard, Kevin ;
Tarabalka, Yuliya ;
Angulo, Jesus ;
Chanussot, Jocelyn ;
Benediktsson, Jon Atli .
IEEE TRANSACTIONS ON IMAGE PROCESSING, 2012, 21 (04) :2008-2021
[6]   Hyperspectral Image Classification With Markov Random Fields and a Convolutional Neural Network [J].
Cao, Xiangyong ;
Zhou, Feng ;
Xu, Lin ;
Meng, Deyu ;
Xu, Zongben ;
Paisley, John .
IEEE TRANSACTIONS ON IMAGE PROCESSING, 2018, 27 (05) :2354-2367
[7]   Towards Evaluating the Robustness of Neural Networks [J].
Carlini, Nicholas ;
Wagner, David .
2017 IEEE SYMPOSIUM ON SECURITY AND PRIVACY (SP), 2017, :39-57
[8]  
Chakraborty A, 2018, ARXIV PREPRINT ARXIV
[9]  
Chen L., 2019, ADVERSARIAL EXAMPLE
[10]   Deep Feature Extraction and Classification of Hyperspectral Images Based on Convolutional Neural Networks [J].
Chen, Yushi ;
Jiang, Hanlu ;
Li, Chunyang ;
Jia, Xiuping ;
Ghamisi, Pedram .
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, 2016, 54 (10) :6232-6251