Mahalanobis masking: a method for the sensitivity analysis of anomaly detection algorithms for hyperspectral imagery

The comparison of anomaly detection capabilities within hyperspectral imagery (HSI) often relies on anecdotal evidence from a small collection of truth-labeled images. This can lead to situations where the performance may not extend to real-world operating conditions. To partially remedy this, we pr...

Full description

Saved in:
Bibliographic Details
Published in:Journal of applied remote sensing 2018-04, Vol.12 (2), p.025001-025001
Main Authors: Messer, Adam J, Bauer, Kenneth W
Format: Article
Language:English
Citations: Items that this one cites
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The comparison of anomaly detection capabilities within hyperspectral imagery (HSI) often relies on anecdotal evidence from a small collection of truth-labeled images. This can lead to situations where the performance may not extend to real-world operating conditions. To partially remedy this, we propose an approach for measuring the sensitivity of an algorithm to the characteristics of the anomalies present. This approach, known as Mahalanobis masking, creates incrementally more difficult to detect anomalies by decreasing their Mahalanobis distance from the background of the image. The sensitivity of an algorithm to changes in the observed anomalies extends the performance evaluation beyond the limited available truth-labeled images. We present both qualitative and quantitative approaches to algorithm evaluation. In a case study, the value of exposing algorithms to a wider range of anomalies is displayed highlighting previously undiscovered algorithm instabilities.
ISSN:1931-3195
1931-3195
DOI:10.1117/1.JRS.12.025001