On the Defense Against Adversarial Examples Beyond the Visible - - PowerPoint PPT Presentation

▶
on the defense against adversarial examples beyond the
SMART_READER_LITE
LIVE PREVIEW

On the Defense Against Adversarial Examples Beyond the Visible - - PowerPoint PPT Presentation

Mar 10, 2024 263 likes •463 views

On the Defense Against Adversarial Examples Beyond the Visible Spectrum Anthony Ortiz 1 , Olac Fuentes 1 , Dalton Rosario 2 , Christopher Kiekintveld 1 1 Department of Computer Science, UTEP 2 US Army Research Laboratory Adversarial Examples on

slide-1
SLIDE 1

On the Defense Against Adversarial Examples Beyond the Visible Spectrum

Anthony Ortiz1, Olac Fuentes1, Dalton Rosario2, Christopher Kiekintveld1

1Department of Computer Science, UTEP 2US Army Research Laboratory

slide-2
SLIDE 2

Adversarial Examples on Natural Images

2

slide-3
SLIDE 3

Adversarial Examples Beyond the Visible Spectrum

3

slide-4
SLIDE 4

Experimantal Setup

4

  • DSTL Dataset:
  • 1 km x 1 km Satellite Image
  • Spatial resolution: 31 cm
  • 3 channels RGB
  • 8 Channels VNIR
  • 8 Channels SWIR
  • 10 Classes (Buildings, roads, track, trees, crops)
  • DigitalGlobe’s WorldView Satellite System
  • Task: Semantic Segmentation
  • Evaluation Metric: Mean IoU
  • Architecture:
  • Fully Convolutional Networks (FCN-8) with VGG-19 as backbone
slide-5
SLIDE 5

Performance Evaluation DSTL Dataset

5

slide-6
SLIDE 6

Adversarial Examples Beyond Visible Spectrum

6

slide-7
SLIDE 7

Adversarial Examples Beyond Visible Spectrum

7

slide-8
SLIDE 8

Adversarial Examples Beyond Visible Spectrum

8

slide-9
SLIDE 9

Dynamic Adversarial Perturbation Attack

9

True Color Input Prediction Clean Prediction Adversarial

slide-10
SLIDE 10

ILFS as a Defense Against Adversarial Examples

10

slide-11
SLIDE 11

Detecting Adversarial Examples

11

slide-12
SLIDE 12

Spectral Signature Adversarial Examples

12

FGSM Iterative FGSM

slide-13
SLIDE 13

Wetness Index

13

Band swir2:1550-1590nm Band swir4: 1710-1750nm

slide-14
SLIDE 14

Detector Network Architecture

14

slide-15
SLIDE 15

Detection Results

15

slide-16
SLIDE 16

Adversarial Training Helps

16

slide-17
SLIDE 17

Adversarial Training Helps

17

slide-18
SLIDE 18

Conclusions

18

  • Multispectral and Hyperspectral Images are vulnerable to adversarial

examples.

  • With the right prior, adversarial examples can successfully be detected.
  • Adversarial Training improve models robustness beyond RGB and generalize

across attacks.

slide-19
SLIDE 19

Thank you

19