적대적 회피 공격에 대응하는 안전한 자율주행 자동차 시스템
Secure Self-Driving Car System Resistant to the Adversarial Evasion Attacks
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
최근 자율주행 자동차는 운전자 지원 시스템에 딥러닝 기술을 적용하여 운전자에게 편의성을 제공하고 있지만, 딥러닝 기술이 적대적 회피 공격(adversarial evasion attacks)에 취약함이 밝혀졌다. 본 논문에서는 객체 인식 알고리즘인 YOLOv5(You Only Look Once)를 대상으로 MI-FGSM (Momentum Iterative-Fast Gradient Sign Method)를 포함한 5가지 적대적 회피 공격을 수행하였으며 객체 탐지 성능을 mAP(mean Average Precision)로 측정하였다. 특히, 본 논문에서는 모폴로지 연산을 적용하여 적대적 공격으로부터 노이즈를 제거하고 경계선을 추출하여 YOLO가 객체를 정상적 탐지할 수 있는 방안을 제안하고 이를 실험을 통해 그 성능을 분석하였다. 실험 결과, 적대적 공격을 수행했을 때 YOLO의 mAP가 최소 7.9%까지 떨어져 YOLO가 객체를 정확하게 탐지하지 못하는 것을 87.3%까지 성능을 개선하였다.
Abstract
Recently, a self-driving car have applied deep learning technology to advanced driver assistance system can provide convenience to drivers, but it is shown deep that learning technology is vulnerable to adversarial evasion attacks. In this paper, we performed five adversarial evasion attacks, including MI-FGSM(Momentum Iterative-Fast Gradient Sign Method), targeting the object detection algorithm YOLOv5 (You Only Look Once), and measured the object detection performance in terms of mAP(mean Average Precision). In particular, we present a method applying morphology operations for YOLO to detect objects normally by removing noise and extracting boundary. As a result of analyzing its performance through experiments, when an adversarial attack was performed, YOLO's mAP dropped by at least 7.9%. The YOLO applied our proposed method can detect objects up to 87.3% of mAP performance.
참고문헌 (22)
- I. Goodfellow, "Explaining and Harnessing Adversarial Examples", International Conference on Learning Representations(ICLR’15), pp.7~9 (2015)
- N. Akhtar, "Threat of adversarial attacks on deep learning in computer vision : A survey", IEEE Access, 6, pp.14410~14430 (2018)
- N. Carlini, "Towards evaluating the robustness of neural networks", IEEE Symposium on Security and Privacy(SP’17), pp.39~57 (2017)
- W. Zhao, "Adversarial Training Methods for Deep Learning : A Systematic Review", Journal of Algorithms, 15(8(283)) (2022)
- N. Papernot, "Distillation as a defense to adversarial perturbations against deep neural networks", IEEE Symposium on Security and Privacy(SP’16), pp.582~597 (2016)
- S. Zheng, "Improving the robustness of deep neural networks via stability training", IEEE Conference on Computer Vision and Pattern Recognition(CVPR’16), pp.4480~4488 (2016)
- W. Xu, "Feature squeezing : Detecting adversarial examples in deep neural networks", Network and Distributed System Security Symposium(NDSS’18), pp.1~15 (2018)
- 이용환, "객체 검출을 위한 CNN과 YOLO 성능 비교 실험", 반도체디스플레이기술학회지, 19(1), pp.85~92 (2020), 한국반도체디스플레이기술학회
- J. Redmon, "You only look once : Unified, real-time object detection", IEEE Conference on Computer Vision and Pattern Recognition(CVPR’16), pp.779~788 (2016)
- Y. Zhang, "Real-Time Vehicle Detection Based on Improved YOLOv5", J. of Sustainability, 14(19), pp.1~19 (2022)
- T. Mostafam, "Occluded Object Detection for Autonomous Vehicles Employing YOLOv5, YOLOX and Faster R-CNN", Proceedings of the IEEE conference on Information Technology, Electronics and Mobile Communication Conference, pp.0405~0410 (2022)
- Y. Dong, "Boosting adversarial attacks with momentum", Proceedings of the IEEE conference on computer vision and pattern recognition, pp.9185~9193 (2018)
- A. Madry, "Towards deep learning models resistant to adversarial attacks", International Conference on Learning Representations(ICLR’18) (2018)
- S. Moosavi-Dezfooli, "Deepfool : a simple and accurate method to fool deep neural networks", IEEE Conference on Computer Vision and Pattern Recognition(CVPR’16), pp.2574~2582 (2016)
- X. Liu, "ADV-BNN : Improved adversarial defense through robust Bayesian neural network", International Conference on Learning Representations(ICLR’19), pp.1~13 (2019)
- A. Athalye, "Obfuscated gradients give a false sense of security : Circumventing defenses to adversarial examples", Proceedings of the 35th International Conference on Machine Learning(ICML’18), pp.274~283 (2018)
- Akhtar, "Threat of adversarial attacks on deep learning in computer vision", Proceedings of the IEEE conference, 6, pp.14410~14430 (2018)
- P. Samangouei, "Defense-gan : Protecting classifiers against adversarial attacks using generative models", Proceedings of ICLR, pp.1~17 (2018)
- M. Dongyu, "Magnet : two-pronged defense against adversarial examples", Proceedings of the ACM SIGSAC, pp.135~147 (2017)
- A. N. Evans, "A morphological gradient approach to color edge detection", IEEE Trans. on Image Processing, 15(6), pp.1454~1463 (2006)
- T. U. Lin, "Microsoft coco : Common objects in context", European Conference on Computer Vision, 8693, pp.740~755 (2014)
- J. Choi, "Adversarial attack and defense of YOLO detectors in autonomous driving scenarios", IEEE Intelligent Vehicles Symposium(IV), pp.1011~1017 (2022)