Virtual Event
March 8 - 12, 2021


Welcome to your Pittcon 2021 Exhibitor Console!

The Exhibitor Console is your hub for all the information you need to know about exhibiting at Pittcon 2021!

Event Information Quick Links
List of Pittcon 2021 Exhibitors Exhibitor Success & ROI Center
Virtual Terms & Condition and Policies
Virtual Pittcon 2021

Targeted Anomaly Detection in Hyperspectral Imaging for Cultural Heritage Applications – Finding a Needle in a Haystack

  • Session Number: S06-04
Friday, March 12, 2021: 10:30 AM - 11:05 AM


Neal Gallagher
Eigenvector Research, Inc.


Hyperspectral imaging is an important analytical tool in a growing number of areas including cultural heritage, forensics, chemical sciences, remote sensing and standoff sensing. Often it is of interest to find minor signal within hyperspectral images for an analyte with a known spectrum i.e., a target. Targeted anomaly detection is defined as the synergistic application of target detection and anomaly detection to achieve a more sensitive and image-specific detection of targets of interest. Target detection uses known targets, often library spectra, to find signal in a hyperspectral image that matches the known signal. Techniques such as matched filter [generalized least squares (GLS)] are often used for target detection. In contrast, anomaly detection is considered ‘non-targeted’ and instead finds unusual signal in an image using empirical techniques such as principal components analysis (PCA). Targeted anomaly detection utilizes GLS weighting strategies with PCA in a weighted principal components analysis (WPCA) that highlights the information of interest. It has been shown that the math of GLS and WPCA are similar but the WPCA version is applied to signal from an image of interest and is thus image-specific and can capture multiple forms of the target signal. The idea behind targeted anomaly detection is to first find pixels in an image that best match a known target using GLS followed by using the detected pixels in the image as target with WPCA. The strategy improves detection sensitivity by a) utilizing target signal relevant for the image under investigation and b) removing detected pixels iteratively to maximize ratio of target-to-clutter signal. After introducing the techniques, examples will be shown for chemical imaging in cultural heritage samples and for stand-off sensing.

Additional Info

Keywords: Please select up to 4 keywords ONLY:
Data Analysis and Manipulation,Spectroscopy,Statistics

For Technical Support with this webpage, please contact support.