Non-literal pattern recognition method and system for hyperspectral imagery exploitation

• US 6,075,891 A
• Filed: 07/06/1998
• Issued: 06/13/2000
• Est. Priority Date: 07/06/1998
• Status: Expired due to Fees

• Current Assignee: General Dynamics Government Systems Corporation (General Dynamics Corporation)
• Original Assignee: General Dynamics Government Systems Corporation (General Dynamics Corporation)
• Inventors: Burman, Jerome Allen
• Primary Examiner(s): Tran, Phuoc
• Assistant Examiner(s): Mariam, Daniel G.
• Application Number:
  US09/111,039
• Time in Patent Office: 708 Days
• Field of Search: 382/190, 382/191, 382/195, 382/284, 382/224, 382/225, 235/454, 235/457
• US Class Current: 382/191
• CPC Class Codes: G06T 2207/20072 Graph-based image processing
• CPC Class Codes: G06T 2207/20084 Artificial neural networks [ANN]
• CPC Class Codes: G06T 7/44 using image operators, e.g. filters, edge density metrics or local histograms
• CPC Class Codes: G06V 20/13 Satellite images

Abstract

A processing system and method identifies materials present in a pixel of a multispectral or a hyperspectral image. Underlying data models representing spectral endmembers or materials of interest and model based relationships are used to identify the materials. The processing system and method provide a technique that detects, unmixes, and classifies spectral signatures for terrain categorization and object identification.

Claims

1. An image processing method, comprising the steps of:

receiving a spectral image of an object;

extracting an input pixel from the spectral image;

converting the input pixel into a vector, each element in the vector of the input pixel representing a reflectance of a respective one of a plurality of spectral bands;

providing spectral signatures of a plurality of endmembers;

correlating a plurality of test spectral images with the spectral signatures of the endmembers;

creating a map of the spectral signatures of the endmembers based on the correlations with the plurality of test vectors;

identifying the spectral signatures of endmembers having a correlation with the input pixel that exceeds a predetermined threshold, including the substep of applying the input pixel to the map to determine which spectral signatures of the endmembers have a correlation with the input pixel that exceeds the predetermined threshold; and

determining a combination of the endmembers present in the input pixel based on the identified endmembers.

2. An image processing method, comprising the steps of:

receiving a spectral image of an object;

extracting an input pixel from the spectral image;

converting the input pixel into a vector, each element in the vector of the input pixel representing a reflectance of a respective one of a plurality of spectral bands;

providing spectral signatures of a plurality of endmembers;

identifying the spectral signatures of endmembers having a correlation with the input pixel that exceeds a predetermined threshold; and

determining a combination of the endmembers present in the input pixel based on the identified endmembers, which further comprises the substeps of;

selecting a group of candidate vectors, each element in the candidate vector representing an abundance of a respective one of the plurality of endmembers;

evaluating each candidate vector with a cost function, which is based upon a constrained spectral mixing model and the identified endmembers; and

determining which candidate vector has the lowest cost, wherein the amounts of the endmembers present in the input pixel corresponds to the elements in the candidate vector determined to have the lowest cost.

3. A method for processing an input pixel of a spectral image, comprising the steps of:

providing spectral signatures of a plurality of endmembers;

correlating a plurality of test spectral images with the spectral signatures of the endmembers;

creating a map of the spectral signatures of the endmembers based on the correlations with the plurality of test vectors;

identifying which spectral signatures of the endmembers have a minimum correlation with the input pixel by applying the input pixel to the map; and

determining a combination of the endmembers present in the input pixel based on the identified endmembers, wherein the determining step includes the substeps of;

selecting a group of candidate vectors, each element in the candidate vector representing an abundance of a respective one of the plurality of endmembers;

evaluating each candidate vector with a cost function, which is based upon a constrained spectral mixing model and the identified endmembers; and

determining which candidate vector has the lowest cost, wherein the amounts of the endmembers present in the input pixel corresponds to the elements in the candidate vector determined to have the, lowest cost.

4. A method for processing an input pixel of a spectral image, comprising the steps of:

scanning an object area to produce the spectral image;

extracting the input pixel from the spectral image;

converting the input pixel into a vector, each element in the vector of the input pixel representing a reflectance of a respective one of a plurality of spectral bands;

providing spectral signatures of a plurality of endmembers;

correlating a plurality of test spectra images with the spectral signatures of the endmembers; and

creating a map of the spectral signatures of the endmembers based on the correlations with the plurality of test vectors;

identifying the spectral signatures of endmembers having a correlation with the input pixel that exceeds a predetermined threshold;

including the substep of applying the input pixel to the map to determine which spectral signatures of the endmembers have a correlation with the input pixel that exceeds the predetermined threshold;

selecting a group of candidate vectors, each element in the candidate vector representing an abundance of a respective one of the plurality of endmembers;

evaluating each candidate vector with a cost function, which is based upon a constrained spectral mixing model and the identified endmembers; and

determining which candidate vector has the lowest cost, wherein the amounts of the endmembers present in the input pixel corresponds to the elements in the candidate vector determined to have the lowest cost.

5. An image processing apparatus comprising:

an input processing section for scanning an object area to produce a spectral image, extracting an input pixel from the spectral image, and converting the input pixel into a vector, each element in the vector of the input pixel representing a reflectance of a respective one of a plurality of spectral bands;

a training data section for providing spectral signatures of a plurality of endmembers;

a material categorization section for identifying the spectral signatures of endmembers having a correlation with the input pixel that exceeds a predetermined threshold, wherein the material categorization section includes subsections for correlating a plurality of test spectral images with the spectral signatures of the endmembers and creating a map of the spectral signatures of the endmembers based on the correlations with the plurality of test vectors, wherein the material categorization section applies the input pixel to the map to determine which spectral signatures of the endmembers have a correlation with the input pixel that exceeds the predetermined threshold; and

a spectral unmixing section for determining a combination of the endmembers present in the input pixel based on the identified endmembers.

6. An image processing apparatus comprising:

an input processing section for scanning an object area to produce a spectral image, extracting an input pixel from the spectral image, and converting the input pixel into a vector, each element in the vector of the input pixel representing a reflectance of a respective one of a plurality of spectral bands,

a training data section for providing spectral signatures of a plurality of endmembers;

a material categorization section for identifying the spectral signatures of endmembers having a correlation with the input pixel that exceeds a predetermined threshold; and

a spectral unmixing section for determining a combination of the endmembers present in the input pixel based on the identified endmembers, further comprising;

a subsection for selecting a group of candidate vectors, each element in the candidate vector representing an abundance of a respective one of the plurality of endmembers;

a subsection for evaluating each candidate vector with a cost function, which is based upon a constrained spectral mixing model and the identified endmembers; and

a subsection for determining which candidate vector has the lowest cost, wherein the amounts of the endmembers present in the input pixel corresponds to the elements in the candidate vector determined to have the lowest cost.