The Science, and Art, of Hyperspectral Image Analysis
Harris Geospatial Solutions – June 7, 2018
Automation is being touted as the wave of the future for analyzing geospatial imagery. But not all data types are created, or can be treated, equally. Hyperspectral analysis has come a long way toward producing automated capabilities however challenges remain. “The promise of automation around hyperspectral has been lauded for years, but there is still enough uncertainty in many cases, that the results need to be corroborated with other information,” explains Gregory Terrie, Principal Scientist at Harris Geospatial Solutions. “There is no guarantee that automated results are 100 percent accurate. No one can say that. Not even us. And ENVI is the recognized industry standard image processing package for hyperspectral analysis.”
Some hyperspectral analysis applications are very challenging, and ENVI lets the user test and evaluate algorithms and approaches. “ENVI was developed specifically to handle hyperspectral data from the ground up, so it is ideal for this,” says Terrie. “You can view the signatures in multiple ways to assess whether these signatures are separable from other things.”
“Looking at an image the way we traditionally look at panchromatic and multispectral data is not where the info is found in a hyperspectral scene, it’s in the full spectra of each pixel,” says Terrie. “ENVI is geared toward handling that efficiently – other tools are not,” Terrie explains.
A hyperspectral scene typically has 200-plus bands and most software packages only support display of either a single band or a red/green/blue color composite. ENVI supports that as well, along with additional ways for visualization and exploration of information contained in the individual pixels. “ENVI is a full-featured image processing package, but has specific tools to handle hyperspectral data including spectral detection and identification that are just not found in other software packages,” says Terrie. Not only are there algorithms for detection and identification, but there are also user interactive capabilities that are only found in ENVI. “There is spectral library support and workflows that are needed to process this data,” explains Terrie. “The processing algorithms in ENVI that are geared toward hyperspectral imagery have been tested and ultimately proven in the community for many, many years.” And where ENVI doesn’t have the exact algorithms that a user needs, the flexibility of IDL and the ENVI API enables users to develop custom routines and applications that can then be integrated with the ENVI application.
For more information about how hyperspectral imagery is being increasingly used to answer important questions in many industries, check out Aaron Andrews blog, Hyperspectral Imaging: An emerging tool for mission readiness. To learn how Harris is making advanced analytics like those that work with hyperspectral data available to the GIS community, be sure to read Jason Wolfe’s blog, Unleashing the power of advanced analytics in ArcGIS.
The real information in hyperspectral data is not in the visible imagery but the individual spectral signatures at each pixel as shown in this image. The peaks and valleys in each signature are characteristic of the material(s) in the pixel.