Presenter Information

Kyle BratcherFollow

Location

https://www.kennesaw.edu/ccse/events/computing-showcase/sp24-cday-program.php

Streaming Media

Document Type

Event

Start Date

25-4-2024 4:00 PM

Description

Advancements in the field of machine learning have led to object detection systems that can approach or even improve upon human performance. Based on deep learning, these systems play a crucial role in many aspects, and continue to be improved on and see expanded adoption. However, these systems are vulnerable to adversarial attacks that rely on targeted noise to spoof detection. Researchers have applied this concept to increase real world adversarial performance by restricting this noise to a patch that can be placed on new images to disrupt object detection. Previous research has focused on patches applied to person recognition (Thys et al. 2019). We focus on the vulnerabilities inherent in remote sensing systems. We develop an adversarial patch to defeat a modern object detection system, YOLOv8. This work demonstrates that remote sensing can still be defeated by an adversarial patch attack and will inform future efforts to develop model robustness against these attacks.

Share

COinS
 
Apr 25th, 4:00 PM

UR-87 Adversarial Patch Attack in Deep Learning Based Remote Sensing Object Detection Model

https://www.kennesaw.edu/ccse/events/computing-showcase/sp24-cday-program.php

Advancements in the field of machine learning have led to object detection systems that can approach or even improve upon human performance. Based on deep learning, these systems play a crucial role in many aspects, and continue to be improved on and see expanded adoption. However, these systems are vulnerable to adversarial attacks that rely on targeted noise to spoof detection. Researchers have applied this concept to increase real world adversarial performance by restricting this noise to a patch that can be placed on new images to disrupt object detection. Previous research has focused on patches applied to person recognition (Thys et al. 2019). We focus on the vulnerabilities inherent in remote sensing systems. We develop an adversarial patch to defeat a modern object detection system, YOLOv8. This work demonstrates that remote sensing can still be defeated by an adversarial patch attack and will inform future efforts to develop model robustness against these attacks.