Back to results
US 5,333,210 AGranted
Method and system for pattern analysis using a coarse-coded neural network
AssigneeThe United States of America as represented by the Administrator of the National Aeronautics and Space Administration
InventorsLiljana Spirkovska, Max B. Reid
Life of this patent
Estimating the term…
Abstract
A method and system for performing pattern analysis with a neural network coarse-code a pattern to be analyzed so as to form a plurality of sub-patterns collectively defined by data. Each of the sub-patterns comprises sets of sub-pattern data. The neural network includes a plurality of fields, each field being associated with one of the sub-patterns so as to receive the sub-pattern data therefrom. Training and testing by the neural network then proceeds in the usual way, with one modification: the transfer function thresholds the value obtained from summing the weighted products of each field over all sub-patterns associated with each pattern being analyzed by the system.
Claims30
A pattern analysis system comprising:
- means for coarse-coding an input pattern to form a plurality of sub-patterns collectively defining the input pattern, each of said sub-patterns comprising respective sets of sub-pattern data, each sub-pattern having a resolution less than a resolution of said pattern and corresponding to a low-resolution pattern of said entire pattern;
- a neural network having a plurality of fields equal in number to said plurality of sub-patterns, so that each field corresponds to one of said sub-patterns, each field including:
- a plurality of input nodes receiving the set of sub-pattern data from its corresponding sub-pattern;
- at least one summation point where weighted products of predetermined combinations of said input nodes are summed to determine a sub-pattern value;
- and
- interconnections between said input nodes and said at least one summation point defining said products and the weights associated with said products;
- said neural network also including:
- at least one output node, coupled to corresponding summation points from a plurality of said fields, which performs a threshold function on a summation of said sub-pattern values received at said at least one output node to produce an output signal said weights have values obtained by training the network to distinguish between a plurality of different training images which have been assigned respective output signals for said threshold function performed at said at least one output node, whereby said system can determine which of said plurality of training images corresponds to said input pattern by providing said output signal for said input pattern.
5 more elements
A pattern recognition system comprising:
- means for coarse-coding a test pattern to form n sub-patterns collectively defining the test pattern, each of said n sub-patterns comprising M.times.M matrices of sub-pattern pixel data, each sub-pattern having a resolution less than a resolution of said test pattern and corresponding to a low-resolution pattern of said entire test pattern;
- a higher order neural network having n fields, each field corresponding to one of said sub-patterns and having a common architecture including:
- M.times.M input nodes receiving said sub-pattern pixel data from its corresponding sub-pattern;
- at least one summation point where weighted products of predetermined combinations of said input nodes are summed to provide a sub-pattern value;
- and
- interconnections between said input nodes and said at least one summation point defining said products and the weights associated with said products, each summation point corresponding to a common summation point in each of the other fields to define at least one set of common summation points;
- said neural network also including:
- at least one output node, corresponding to a respective one of said at least one set of common summation points, which performs a threshold function on a summation of the sub-pattern values produced by the set of common summation points corresponding to said output node, said weights having values obtained by training said network to distinguish between a plurality of different training patterns which have been assigned respective output signals for said threshold function performed at said at least one output node whereby said system can determine which of said plurality of training images corresponds to said test pattern by providing said output signal for said test pattern.
5 more elements
A method of analyzing a pattern comprising:
- coarse-coding an input pattern to form a plurality of sub-patterns collectively defining the input pattern, each of said sub-patterns comprising respective sets of sub-pattern data, each sub-pattern having a resolution less than a resolution of said pattern and corresponding to a low-resolution pattern of all of said pattern;
- supplying said sub-pattern data to input nodes of a neural network having a plurality of fields equal in number to said plurality of sub-patterns, so that input nodes of each field receive the sub-pattern data from a corresponding one of said sub-patterns;
- within each field, summing weighted products of predetermined combinations of said input nodes at summation points to produce at least one sub-pattern value for each field;
- and
- supplying a summation of the sub-pattern values from a plurality of said fields to an output node of the neural network, coupled to corresponding summation points from a plurality of said fields, said output node performing a threshold function on said summation of sub-pattern values to produce an output signal
- the weights associated with said products have values obtained by training the network to distinguish between a plurality of different training images which have been assigned respective output signals for said threshold function performed at said at least one output node, whereby said system can determine which of said plurality of training images corresponds to said input pattern by providing said output signal for said input pattern.
3 more elements
Dependencies are inferred from the claim text.