|
Activity Number:
|
342
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Tuesday, August 2, 2016 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Statistics in Defense and National Security
|
|
Abstract #319311
|
View Presentation
|
|
Title:
|
Hyperspectral Video Analysis Using Graph-Clustering Methods
|
|
Author(s):
|
Zhaoyi Meng*
|
|
Companies:
|
|
|
Keywords:
|
multi-class classification ;
hyperspectral video ;
energy minimization ;
MBO scheme ;
Nystrom Extension Method ;
parallel computing
|
|
Abstract:
|
Perhaps the most challenging imaging modality to analyze in terms of the vast size of the data are videos taken from hyperspectral cameras. We consider an example involving standoff detection of a gas plume involving Long Wave Infrared spectral data with 128 bands. Rather than using PCA or a similar dimension reduction method we treat this as a "big data" classification problem and simultaneously process all pixels in the entire video using novel new graph clustering techniques. Computation of the entire similarity graph is prohibitive for such data so we use the Nystrom extension to randomly sample the graph and compute a modest number of eigenfunctions of the graph Laplacian. A very small part of the spectrum allows for spectral clustering of the data. However with a larger but still modest number of eigenfunctions we can solve a graph-cut based semi supervised or unsupervised machine learning problem to sort the pixels into classes. We discuss challenges of running such code on both desktops and supercompers.
|
Authors who are presenting talks have a * after their name.
Back to the full JSM 2016 program
|
