Image analysis – Pattern recognition – Feature extraction
Reexamination Certificate
1999-07-12
2003-12-16
Au, Amelia M. (Department: 2623)
Image analysis
Pattern recognition
Feature extraction
C382S260000, C382S276000
Reexamination Certificate
active
06665438
ABSTRACT:
BACKGROUND OF THE PRESENT INVENTION
1. Field of the Invention
The present invention relates to remote sensing imagery processing, and more particularly to a method for hyperspectral imagery exploitation and unmixing using hybrid approaches including evolutionary computing techniques and robust filtering techniques.
2. Description of Related Arts
The term, hyperspectral data, is ambiguous in and of itself The unifying trait of all hyperspectral data is the existence of a gross quantity of specific and minuscule spectral bands located within the optical wavelength region. The exact quantity of bands relating to any hyperspectral image varies widely. A single band range located within the visible wavelength region might vary between a single nanometer to hundreds of nanometers. Band ranges located within the infrared and thermal wavelength regions might exceed those for the visible wavelength region. Of course, hyperspectral data is exceedingly desirable, due to the ease with which one may recognize observed entities based on very specific characteristic features, corresponding to those entities, which are associated with very narrow spectral bands. This type of detection and recognition is simply not possible using traditional methods.
The disadvantage associated with hyperspectral data is the necessary capability to process extraordinary amounts of information. Specific elements, entities or objects, or components thereof, possess specific spectral signatures. One's ability to ascertain a specific spectral signature results in one's ability to ascertain the corresponding element. Previous methods for dealing with hyperspectral data included pattern matching techniques. These techniques rely upon models and least squares algorithms in order to recognize and isolate elements within hyperspectral data. These pattern matching techniques are limited by their lack of robustness. Their results degrade significantly across spatial and temporal variations. They are inadequate at recognizing elemental components from a combined spectral signature. They require a tremendous amount of computation. Their results degrade significantly across sensor and atmospheric variations. They do not deal with nonlinearity well. Also, these techniques do not respond well to increasing databases of elements for which to detect within the hyperspectral data.
Evolutionary Computing (EC) techniques lend themselves well to nonlinear optimization. Evolutionary Computing techniques result from an aggregate of recently developed sophisticated computational fields, such as Genetic Algorithms, Genetic Programming, Artificial Neural Networks, and Artificial Life. These realms are ideal for sifting through hyperspectral data and performing pattern matching. Aside from matching spectral signatures compiled in databases to spectral signatures extracted from hyperspectral imagery, these EC techniques may yield the very procedures, resulting from iterations of searches, which may be used to analyze the hyperspectral data to begin with. These techniques have the benefit of “learning.” The rudimentary basis of an Evolutionary Algorithm (EA) is a pool of arbitrary entities, wherein each entity represents a method for analyzing a specific aspect of a set of data. Please refer to
FIG. 1
for a depiction of an Evolutionary Algorithm.
Darwin's survival of the fittest concept of genetics and evolutionary processes serves as the framework for the Evolutionary Algorithmic approach to optimization tasks. Beginning with a pool of possible candidate methods for remedying a certain problem, a new generation of unique, and supposedly better (i.e. more fit), methods results from arbitrary pairings, or matings, if you will, of the parent, or original, generation of methods. Once the parent methods have been paired off to mate, they swap parts of their individual methods with one another, This exchange of method, or chromosomal information, is called crossover or reproduction. Two distinct, original methods result from the reproduction process and the parent methods no longer exist. Only the offspring survive the crossover process. The offspring must be capable of analyzing the same aspects of the data as their parents were able to do so. Mutations occur arbitrarily within each successive generation to introduce necessary diversity.
The offspring methods of each successive generation are tested as to their abilities in analyzing the data. The user stipulates the criteria against which the offspring methods are judged. The methods of each generation receive scores. These scores designate the fitness of each method, i.e. the appropriateness of each method in analyzing the data according to the user's criteria. Less fit methods are not allowed to mate and produce successive generations, therefore they are extinguished and become extinct. Methods, which have achieved high fitness scores, are allowed to mate and crossover, therefore they propagate themselves and continue their species. This survival of the fittest evolutionary scheme continues for a specific, user selected, number of generations. This approach results in a swift convergence upon the optimal method for analyzing a given pool of data. Tasks, such as spectral unmixing and object detection and identification lend themselves exceedingly well to this type of evolutionary approach. Therefore, as an extension of the above statement, hyperspectral data present themselves as ideal candidates for Evolutionary Computing techniques.
The resulting method with the best fitness score, after having been judged against the user's criteria, at the end of the process, after all of the allowed generations have propagated themselves, is the optimal method for analyzing the given set of data. Evolutionary Algorithms typically provide much better solutions than traditional methods, due to the fact that the search process is possesses much greater breadth of options. Evolutionary Algorithms conduct a simultaneous scouring of the entire pool of candidates. Evolutionary Algorithms do not choke on complex optimization tasks, including constraints. Evolutionary Algorithms have no problem with nonlinearities. Nonlinearities often present themselves in relation to atmospheric and sensor constraints associated with hyperspectral imagery data. The cost function associated with the resulting fitness judgments reflect these nonlinearities and constraints.
FIG. 2
represents a depiction of the application of Evolutionary Computing techniques to hyperspectral data. This application includes a reliable technique for processing distinct pixel signatures in relation to spectral wavelength. Evolutionary Algorithm techniques include considerations of signature preprocessing and non parametric searching according to model based constraints as applied to hyperspectral imagery data.
Tens of thousands of iterations are required for each and every pixel of a hyperspectral image when applying Evolutionary Computing techniques. This necessity results in a very slow computation process. A reasonably sized hyperspectral image might require millions and millions of iterations. Therefore a fast process for hyperspectral imagery exploitation and accurate unmixing is required.
SUMMARY OF THE PRESENT INVENTION
A main objective of the present invention is to provide a hybrid approach which uses robust filtering techniques to perform fast pixel unmixing with hyperspectral imagery, and genetic algorithm so as to fine the abundance estimation derived by the robust filter.
Another objective of the present invention is to provide a hybrid approach which uses robust filtering techniques to perform fast pixel unmixing with hyperspectral imagery, and genetic algorithm so as to derive accurate abundance estimation when the estimation error of the robust filter is larger than a preset threshold.
Another objective of the present invention is to provide a hybrid approach which uses robust Kalman filter to perform fast pixel unmixing with hyperspectral imagery, and genetic algorithm to fine the abundance estimation de
American GNC Corporation
Au Amelia M.
Chan Raymond Y.
David and Raymond Patent Group
Le Brian
LandOfFree
Method for hyperspectral imagery exploitation and pixel... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method for hyperspectral imagery exploitation and pixel..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method for hyperspectral imagery exploitation and pixel... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3140042