Method and apparatus for predicting object properties and events using similarity-based information retrieval and modeling
Inventors
Birdwell, J. Douglas • Wang, Tse-Wei • Icove, David J. • Horn, Sally P. • Horn, Roger • Rader, Mark S. • Stansberry, Dale V.
Assignees
University of Tennessee Research Foundation • United States Department of the Army
Publication Number
US-8396870-B2
Publication Date
2013-03-12
Expiration Date
2030-06-25
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 predicting properties of a target object comprise application of a search manager for analyzing parameters of a plurality of databases for a plurality of objects, the databases comprising an electrical, electromagnetic, acoustic spectral database (ESD), a micro-body assemblage database (MAD) and a database of image data whereby the databases store data objects containing identifying features, source information and information on site properties and context including time and frequency varying data. The method comprises application of multivariate statistical analysis and principal component analysis in combination with content-based image retrieval for providing two-dimensional attributes of three dimensional objects, for example, via preferential image segmentation using a tree of shapes and to predict further properties of objects by means of k-means clustering and related methods. By way of example, a fire event and residual objects may be located and qualified such that, for example, properties of the residual objects may be qualified, for example, via black body radiation and micro-body databases including charcoal assemblages.
Core Innovation
The invention relates to a method and apparatus for predicting properties of target objects through similarity-based information retrieval and modeling by analyzing parameters from multiple databases. These databases include an electrical, electromagnetic, acoustic spectral database (ESD), a micro-body assemblage database (MAD), and an image data database. The method employs multivariate statistical analysis and principal component analysis combined with content-based image retrieval to provide attributes of three-dimensional objects, such as using preferential image segmentation via a tree of shapes. Properties of objects can then be predicted through clustering methods like k-means, enabling qualification and location of events such as fires and residual objects by analyzing characteristics like black body radiation and micro-body assemblages.
The background identifies problems in existing database systems, search and retrieval methods for visual image databases, cluster generation in multi-dimensional concept spaces, and regression-based qualitative analysis. Known methods often require exact matches or rely on passive indexing without integrating multiple data sources and complex object properties. There is a need for an automated and efficient method to infer object properties and contextual information when exact matches are unavailable by utilizing data from multiple reference databases with statistical models and similarity measures.
Claims Coverage
The patent claims cover methods and systems involving multiple specialized databases and modeling techniques for predicting properties of objects based on similarity measures.
Integration of micro-body assemblage and frequency spectral databases with a processor search manager apparatus
The system stores data objects in micro-body assemblage and frequency spectral databases including micro-body assemblage data types (pollen, charcoal, diatom, foraminifera) associated with geographic or environmental properties, and frequency spectral data that include electromagnetic spectra with associated attributes.
Similarity-based search and retrieval for target object properties
A search manager apparatus processes a query containing micro-body or frequency spectral data, searches the corresponding database, retrieves information about objects with similar characteristics using similarity measures, and displays the target object and related similar objects visually as linked data points.
Model-based prediction of object properties
Application of models to retrieved data to predict properties of the target object different from the query data, such as geographic location, origin, source, region, or environmental properties.
Multivariate statistical analysis and clustering for indexing and efficient access
Generating indices of databases by analyzing data attributes via multivariate statistical analysis to determine a reduced subspace, clustering data objects based on similarity measures projecting vector attributes onto this subspace to facilitate efficient retrieval of similar objects.
Use of multiple databases with linked evidence trees and graphical user interfaces
Providing interactive displays and evidence trees showing circular-linked representations of objects and associated data across micro-body assemblage, electro-acoustic spectral databases, and image data, differentiated visually and linked by similarity metrics.
Utilization of optimization, statistical and machine learning models
Employing various models including Bayesian, principal component analysis, least squares, maximum likelihood, artificial neural networks, fuzzy logic, and hierarchical data models to predict and infer object properties based on similar database entries.
Dynamic and scalable similarity-based indexing and searching techniques
Implementing tree-structured indices with multivariate statistical partitions and dynamic indexing to provide rapid and scalable similarity searches across large multidimensional datasets.
The claims collectively describe a comprehensive framework combining multiple specialized databases with similarity-based search methods and statistical and machine learning models to efficiently predict properties of objects from diverse data types including micro-body assemblages and frequency spectral data, accompanied by dynamic indexing and interactive visualization tools.
Stated Advantages
Enables rapid selection of objects from large databases based on similarities to a target object, enabling efficient property prediction.
Supports the fusion of multiple data types (electrical, spectral, micro-body assemblage, image data) for enhanced object property inference.
Provides highly scalable and efficient indexing and search methods suitable for very large datasets with demonstrated performance.
Allows dynamic updating of databases and models to incorporate new data, improving prediction accuracy over time.
Uses advanced image segmentation and content-based image retrieval methods to automate and improve micro-particle identification.
Documented Applications
Forensic analysis including identifying geographic origin, source, or environmental properties of objects using trace micro-body assemblages such as pollen, charcoal, diatoms, and foraminifera.
Fire event and combustion residual detection and characterization using electrical, electromagnetic, and acoustic spectral data.
Environmental and climate studies through analysis of micro-fossils and micro-body assemblage distributions.
Identification and classification of micro-particles and micro-bodies via image databases using content-based image retrieval.
Monitoring and tracking of living beings and vehicles by thermal and spectral signature analysis.
Industrial process monitoring and fault detection through similarity-based analysis of time series data.
Financial data mining, criminal activity pattern detection, disease modeling, and drug discovery by similarity-based clustering and modeling of object attributes.
Interested in licensing this patent?