Abstract
We have developed a method for using hyperspectral (HS) data to identify and locate chemical materials on arbitrary surfaces using the materials’ reflection or emission spectra that makes no prior modeling assumptions about the presence of pure pixels or the statistics of the background clutter and sensor noise. To our knowledge, this is the first time that surface detection without dependence on background information has been achieved. There are three main components to the method: (1) an HS unmixing algorithm based on the alternating direction method of multipliers that is applied over local subsets of the imaging to resolve the HS data into a set of linearly independent spectral and spatial components; (2) the fitting of those unmixing spectra to a set of candidate template spectra; and (3) a support vector machine classifier for chemical detection, identification, and location. The algorithm is illustrated on HS data collected by a Telops Hyper-Cam infrared camera on data resulting from the deposition of chemical agent simulants on various surfaces.
Highlights
Hyperspectral (HS) imaging is a well-established technology having many commercial and military applications
Traditional approaches for separating the radiance components in the data involve either (1) unmixing using the “purepixel” or “endmember” assumption that the target materials can be found in a few individual pixels without interference from other radiance sources or (2) the use of likelihood ratio testing in a statistical approach
We have developed an entirely different method for identifying and locating materials on unknown background surfaces using their chemical spectra that makes no prior modeling assumptions about the presence of endmember pixels or the statistics of the surface clutter and sensor noise
Summary
Hyperspectral (HS) imaging is a well-established technology having many commercial and military applications. Warren and Cohn: Chemical detection on surfaces by hyperspectral imaging probability densities The latter method assumes that the clutter correlation structure can be estimated from the data, assuming spatial homogeneity and the ability to mask the target spectral presence. In addition to the need for constraints, we have observed that much better results are obtained by applying the unmixing and spectral fitting to local subsets of the total data cube This is true when the materials of interest are present in relatively small regions of the total imagery. For other applications, such as vapor data with larger plume-targets and higher noise levels, blocks of about 30 pixels on a side are more appropriate These choices allow the processing to adapt to the local structure of the image while supplying enough spatial information to generate good unmixing results. Besides providing better spectral estimates, the use of local processing can be computationally advantageous over a single global fit since the processing steps are identical for each local region and can, be implemented by parallel processing methods
Sign-in/Register to view highlights & summary Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
