Large Spectral Library Problem
Hyperspectral imaging produces a spectrum or vector at each image pixel. These spectra can be used to identify materials present in the image. In some cases, spectral libraries representing atmospheric chemicals or ground materials are available. The challenge is to determine if any of the library chemicals or materials exist in the hyperspectral image. The number of spectra in these libraries can be very large, far exceeding the number of spectral channels collected in the ¯eld. Suppose an image pixel contains a mixture of p spectra from the library. Is it possible to uniquely identify these p spectra? We address this question in this paper and refer to it as the Large Spectral Library (LSL) problem. We show how to determine if unique identi¯cation is possible for any given library. We also show that if p is small compared to the number of spectral channels, it is very likely that unique identi¯cation is possible. We show that unique identi¯cation becomes less likely as p increases.
- Research Organization:
- Pacific Northwest National Lab. (PNNL), Richland, WA (United States)
- Sponsoring Organization:
- USDOE
- DOE Contract Number:
- AC05-76RL01830
- OSTI ID:
- 963240
- Report Number(s):
- PNNL-17810; NN2001000; TRN: US200917%%361
- Country of Publication:
- United States
- Language:
- English
Similar Records
Multivariate analysis of X-ray, ion and electron spectral images: from surface to 3D materials characterization.
Using soil library hyperspectral reflectance and machine learning to predict soil organic carbon: Assessing potential of airborne and spaceborne optical soil sensing