RSC - Research - Spectral Imaging 2

Spectral Imaging

I've been working in this area since 1995 or so, but most of the output is in student theses, either classified or not.  

My first remote sensing student was back in 1994 - Captain Sturgeon worked with AOTF spectral data from JPL. (Spectral and polarimetric analysis of hyperspectral data collected by an acousto-optic tunable filter system., SPIE, 1994.)  Really, she was Phil Durkee's student, but I did most of the software.  This was pre-ENVI days!

HYDICE was started by one of my first thesis students, Mark Landers.  He eventually went off to a company named SpecTir. As part of a fairly broad interest in spectral imaging, we did a number of studies on statistics in spectral data, and showed that the 'Gaussian' assumption people always make is wrong. Statistics of Target Spectra in HSI Scenes with Scott Tyo and students

One of my very best students was Marine Captain Marcus Stefanou.  He did a survey of the existing statistics based analysis techniques for spectral imagery that is still a reference standard for the field. A Signal Processing Perspective of Hyperspectral Imagery Analysis Techniques.  Marcus got fed up with his career path, and moved over to the AF.  He never learned to let his hair grow out, but he made Colonel and got a PhD at Rochester Institute of Technology (RIT).   He recently joined the faculty at NPS.

We did some work on bathymetry using spectral techniques - see that page.

We did a fair amount of classified work on Long Wave Infrared (LWIR) spectral imaging.  Brian Collins (another Marine) had worked with John Hackwell at Aerospace, and did an outstanding job working with SEBASS.  Thermal Spectral Imagery Analysis is the SPIE paper from that work.  His thesis is here. In an attempt to break out of the classified arena, I sent Marine Captain Aimee Mares off to the University of Hawaii, to work with Dr. Paul Lucey.  She brought home some of the first volcano measurements taken in the LWIR. LWIR Spectral Measurements of Volcanic Sulfur Dioxide Plumes.  This would have made a great journal publication, if the wind had simply blown the SO2 plume over the Chain of Craters road, like it does almost every day.  As it was, we didn't get the confirming UV measurements to validate.sad

For a short time I had a pretty nice deal going with Kip Krebs from the OR department, and Scott Tyo from ECE.  Kip got fed up with life at NPS, and left academia completely.  That sort of ended our human factors work.  Scott, an AF Captain, left the AF.  NPS failed to make him a timely offer, so he  moved on to the University of New Mexico, then the University of Arizona, and finally the UNSW Canberra, where he seems to have prospered.  The work with Scott continued on a second topic, and we have a nice paper on invariant display strategies.  Principal-Components-Based Display Strategy for Spectral Imagery

As soon as they came online, it was clear to me that the commercial imagery systems, like Quickbird and IKONOS, were the future of remote sensing.  One of our early studies on spectral imaging with IKONOS was a study of the Elkhorn Slough.  Terrain Classification in Urban Wetlands with High-spatial Resolution Multi-spectral Imagery,

In the latter half of 2004, I had a post-doc, of sorts, from Taiwan.  Brandt Tso came over for a break from his regular duties, and worked terribly hard on spectral imaging, adapting Markov Fields concepts to spectral classification.  He took some unique, multiangle observations I had from Quickbird, and looked at angular (BDRF) and texture effects.  Here is the SPIE talk.   Scene classification using combined spectral, textural and contextual information.    The publications on his Markov random field work while in Monterey are: A contextual classification scheme based on MRF model with improved parameters estimation and multiscale fuzzy line process, and  Combining spectral and spatial information into hidden Markov models for unsupervised image classification

I had an early idea that viewing the earth at high spatial resolution from multiple angles might be useful, and My version of the look at the multi-angle data was given at the same conference - Multilook scene classification with spectral imagery. (again, 4 Quickbird observations in sequence over Fresno, CA. over 4 minutes)


rco, 5/25/2017

Navigation Menu