Population of background suppression lists from limited data in agent detection systems
Methods and systems are disclosed for detection of agents such as pathogens or toxic substances and, in particular, to methods and systems for determining the most important background constituents to suppress in a sample, e.g., in a bulk aerosol sample, in order to reduce the probability of false alarms and improve the level of detection of potentially harmful airborne agents.
1. A method for detecting a target constituent in a mixture of background constituents, comprising
utilizing a measurement modality to interrogate the mixture with electromagnetic radiation so as to generate sample spectral data corresponding thereto,
deriving principal components of the sample spectral data,
applying a rotate-and-suppress (RAS) transformation to said principal components of the sample data to suppress a contribution of at least one background constituent contained in a suppression list of length X, where X is less than the number of independent components used to describe each constituent, and
comparing said transformed principal components of the sample data with background-suppressed principal components of corresponding spectral data of a target constituent to determine whether said target constituent is present in the mixture.
2. The method of claim 1 in which the components of each of said target and background constituent is determined by principal component analysis.
3. The method of claim 1 in which N constituents are chosen from recent data and M constituents are chosen from a background library.
4. The method of claim 3 in which the background library includes measurements of known substances.
5. The method of claim 3 in which the suppression list includes N endmembers from a set of recent data.
6. The method of claim 3 in which the N constituents chosen from recent data pass a threshold test.
7. The method of claim 3 in which the background library includes measurements of pure substances and endmembers determined by analysis of mixture data collected from regions of interest.
8. The method of claim 3 in which the N constituents chosen from recent data and the M constituents chosen from a background library do not suppress any members of an Agent Library below a predetermined threshold using the rotate-and-suppress method.
9. The method of claim 3 in which the N constituents chosen from recent data are the N most recent measurements which did not trigger an alert or an alarm, and the M constituents chosen from the background library are permanent members of the suppression list.
10. The method of claim 9 in which the number of significant endmembers is determined by the distances of the respective endmembers from the simplex defined by the next smaller set of endmembers.
11. The method of claim 3 in which a single constituent chosen from recent data has the minimum spectral angle moment of inertia of any member of the recent data set.
12. The method of claim 11 in which a first constituent chosen from recent data has the maximum spectral angle moment of inertia of any member of the recent data set.
13. The method of claim 3 in which the suppression list includes N endmembers from a set of recent data augmented by the constituents contained in the background library.
14. The method of claim 3 in which the suppression list is the set of M constituents from the background library within a spectral angle threshold of at least one significant recent measurement contained in a set of recent measurements.
15. The method of claim 1 , further comprising
utilizing said measurement modality to obtain spectral data corresponding to at least an agent,
deriving principal components of the agent spectral data, and
applying said rotate-and-suppress transformation to said principal components of the agent data to generate said background-suppressed principal components of the agent data.