Method and apparatus for classifying known specimens and media using spectral properties and identifying unknown specimens and media
Inventors
Birdwell, John Douglas • Sapp, Carl G. • Wang, Tse-Wei • Icove, David J. • Horn, Roger • Rader, Mark S. • Stansberry, Dale V.
Assignees
University of Tennessee Research Foundation • United States Department of the Army
Publication Number
US-8429153-B2
Publication Date
2013-04-23
Expiration Date
2032-05-11
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Abstract
Method and apparatus for determining a metric for use in predicting properties of an unknown specimen belonging to a group of reference specimen electrical devices comprises application of a network analyzer for collecting impedance spectra for the reference specimens and determining centroids and thresholds for the group of reference specimens so that an unknown specimen may be confidently classified as a member of the reference group using the metric. If a trait is stored with the reference group of electrical device specimens, then, the trait may be predictably associated with the unknown specimen along with any traits identified with the unknown specimen associated with the reference group.
Core Innovation
The invention relates to a method and apparatus for classifying known specimens and media using spectral properties and identifying unknown specimens and media having like spectral properties. It involves application of a spectral property database that stores data values representative of spectral properties such as magnitude, phase, or complex values sampled at discrete values of time, frequency, wavelength, energy, or other scalar quantities. Preference is given to the use of one or more metrics for classification and identification, with an evaluation of thirteen different similarity metrics leading to the selection of the five best performing metrics for classification and identification. The method includes collecting spectral data from known specimens and media, storing them in databases along with specimen properties and traits, and then comparing spectral data of unknown specimens or media to the known spectral data to predict properties or traits.
The invention addresses the problem that known methods in database systems and target recognition struggle with representing regions or clusters of spectral data efficiently and with comparing unknown spectra to these regions in a way suitable for large volumes of data. Prior approaches often rely on limited sets of statistics such as centroids but lack effective representations for entire regions or clusters. Additionally, there is a need for a method that can handle spectral data of various kinds, including electrical impedance spectra, black body radiation, acoustic data, and other spectral forms across wide frequency ranges. The disclosed invention provides a framework for efficient classification of spectral data into reference groups and for associating traits of these groups with unknown samples to predict properties or values of traits from the similarity metrics and clustered data.
Claims Coverage
The patent claims cover methods and apparatuses for classifying unknown specimens to reference groups using spectral data by employing selected similarity metrics, dimension reduction, thresholds, and associated trait prediction.
Selection and use of similarity metrics for specimen classification
The selection and application of a similarity metric chosen from a plurality of metrics including Manhattan, Canberra, similarity index, cosine, and Euclidean distance to classify an unknown specimen to a reference group in a spectral database.
Dimension reduction of spectral data
Reducing the dimensionality of input spectral data using techniques such as binning by spectral frequency, principal component analysis, or projection search methods to enable efficient processing for classification.
Classification scenarios for group membership
Employment of one of three classification scenarios: (1) classifying the unknown specimen to the reference group regardless of threshold, (2) classifying based on exceeding a similarity threshold, and (3) maintaining the specimen as unclassified if it meets thresholds for multiple groups.
Prediction and storage of traits based on group membership
Predicting values of traits for an unknown specimen upon membership classification and storing traits of the unknown specimen associated with the reference group in the spectral database.
Use of model analysis for spectral data
Optional fitting of input spectral data to a circuit model such as an equivalent high order polynomial or transfer function for parameter estimation of the unknown specimen.
Database indexing and search using high performance similarity-based methods
Construction and utilization of indexed spectral databases, including vector attribute representation, clustering, dynamic indexing, and search engines to enable rapid retrieval and prediction of properties from large spectral datasets.
Graph-based evidence representation for objects
Representation of objects and associated information as graphs or evidence trees, integrating multiple data modalities and supporting associative data modeling for forensic and classification applications.
The claims encompass a comprehensive computer-implemented method and apparatus that employ dimension reduction, choice of optimized similarity metrics, thresholding, and predictive modeling using spectral data to classify unknown specimens to reference groups and infer their traits within a scalable and efficient database framework.
Stated Advantages
Provides efficient and scalable similarity-based searching and classification of specimens and media using spectral data.
Enables prediction of traits and properties of unknown specimens based on classified reference groups.
Supports use of multiple similarity metrics optimized for the data type and classification scenario.
Allows classification scenarios to balance error tolerance and decision conservatism for application-specific needs.
Integrates multiple data types and measurement modalities to improve accuracy of identification and trait prediction.
Supports dynamic updating and expansion of databases with new specimens and traits for continued improvement of inferences.
Provides a platform for automated analysis, clustering, and indexing facilitating large database handling with rapid response times.
Documented Applications
Classification of electrical devices by impedance spectral data to identify manufacturer, type, or status and predict traits such as temperature or failure state.
Recognition and identification of biological specimens, humans, animals, or musical instruments through voice and spectral analysis.
Detection and characterization of fire events and residual charred remains via black body microwave and acoustic spectral emission.
Geographic and environmental origin identification of specimens using spectral and micro-body assemblage data such as charcoal and pollen.
Forensic applications including trace evidence analysis, linking samples to geographic regions or sources based on spectral and micro-body data.
Medical diagnosis by identification of tissue conditions such as cancer via impedance and spectral analysis.
Financial data mining and pattern detection using multivariate time series analysis and clustering.
Criminal activity pattern detection through analysis of suspicious transaction behaviors and network interactions.
Disease modeling and drug discovery through genetic and phenotypic trait analysis using similarity and clustering methods.
General classification and identification applications involving spectral data of acoustics, optics, electromagnetic radiation, and isotopic measurements.
Interested in licensing this patent?