Rapid detection

摘要

In accordance with various aspects of the disclosure, a detecting engine for detecting targets/materials in hyperspectral scenes is disclosed. The detecting engine combines data partitioning and dimensionality reduction to reduce the number of computations needed to identify in which pixels in a hyperspectral scene a given material is present. Computation reduction (in some instances, by two fold) greatly impacts the speed of and power consumed by the detecting engine making the engine suitable for hyperspectral imaging of large scenes, processing using many filters per pixel, or missions requiring testing large numbers of reference spectra to see which are present in a scene.

主权项

1. A method for detecting materials in a hyperspectral scene, the method comprising:
in a detecting engine provided with: i) a spectral library of references in which each reference is a spectrum of a material, ii) dimensional reduced representations of the spectra of a first pluralities of pixels included in the hyperspectral scene, the representations including basis vectors and dimensional reduced coefficients, and iii) full spectra of a second plurality of pixels included in the hyperspectral scene, with respect to a subject spectrum of a respective material from the spectral library of references; for each pixel in the second plurality of pixels, determining a score indicative of the likelihood that the spectrum of a respective pixel matches the subject spectrum of the respective material; transforming the subject spectrum of the respective material into dimensional reduced library coefficients based on the provided basis vectors; for each pixel in the first plurality of pixels, determining which pixel is a variant pixel having dimensional reduced coefficients sufficiently different from dimensional reduced coefficients associated with other pixels; for each variant pixel, determining a score indicative of the likelihood that the dimensional reduced coefficients associated with a respective variant pixel match the dimensional reduced library coefficients of the respective material; comparing the scores to a threshold; based on the comparison, determining in which of the pixels in the hyperspectral scene the respective material is present; calculating an error associated with the dimensional reduced library coefficients of the respective material; comparing the calculated error to a second threshold; and based on the comparison determining that the respective material is not present in any of the pixels in the hyperspectral scene represented by dimensional reduced coefficients.