A Multi-Scale Isolation Forest Approach for Real-Time Detection and Filtering of FGSM Adversarial Attacks in Video Streams of Autonomous Vehicles

2025-02-22

AI Summary

This paper addresses the challenge of real-time detection of Fast Gradient Sign Method (FGSM) adversarial attacks in autonomous driving video streams. We propose a lightweight, gradient-free, and retraining-free multi-scale Isolation Forest method that extracts multi-resolution image features to model anomalous patterns induced by FGSM perturbations across varying intensities (ε = 0.01–0.2), enabling efficient identification and filtering of adversarial samples. To our knowledge, this is the first work to adapt multi-scale Isolation Forests to real-time video-stream adversarial detection. Evaluated on 10,000 test images under five perturbation levels, our approach achieves >96% detection rate for all levels while significantly reducing misclassification rates. It simultaneously ensures robustness, low-latency inference, and strong generalization across unseen perturbation strengths. The implementation is publicly available.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Machine Learning: Adversarial Learning & Robustness
• Multiagent Systems: Adversarial Agents

Application Category

• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Security and Privacy: Large-scale security measurements
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Deep Neural Networks (DNNs) have demonstrated remarkable success across a wide range of tasks, particularly in fields such as image classification. However, DNNs are highly susceptible to adversarial attacks, where subtle perturbations are introduced to input images, leading to erroneous model outputs. In today's digital era, ensuring the security and integrity of images processed by DNNs is of critical importance. One of the most prominent adversarial attack methods is the Fast Gradient Sign Method (FGSM), which perturbs images in the direction of the loss gradient to deceive the model. This paper presents a novel approach for detecting and filtering FGSM adversarial attacks in image processing tasks. Our proposed method evaluates 10,000 images, each subjected to five different levels of perturbation, characterized by $epsilon$ values of 0.01, 0.02, 0.05, 0.1, and 0.2. These perturbations are applied in the direction of the loss gradient. We demonstrate that our approach effectively filters adversarially perturbed images, mitigating the impact of FGSM attacks. The method is implemented in Python, and the source code is publicly available on GitHub for reproducibility and further research.

Problem

• Detect FGSM adversarial attacks
• Filter perturbed images in real-time
• Secure DNNs in autonomous vehicles

Innovation

• Multi-Scale Isolation Forest
• Real-Time FGSM detection
• Autonomous Vehicles security