Rough wavelet granular space and classification of multispectral remote sensing image

• US 9,152,877 B2
• Filed: 01/13/2011
• Issued: 10/06/2015
• Est. Priority Date: 11/24/2010
• Status: Expired due to Fees

• Current Assignee: Indian Statistical Institute
• Original Assignee: Indian Statistical Institute
• Inventors: Pal, Sankar Kumar, Meher, Saroj Kumar
• Primary Examiner(s): PARK, SOO JIN
• Application Number:
  US13/265,325
• Publication Number:
  US 20120183225A1
• Time in Patent Office: 1,727 Days
• Field of Search: None
• US Class Current: 1/1
• CPC Class Codes: G06T 2207/10036 Multispectral image; Hyperspectral image
• CPC Class Codes: G06T 2207/20064 Wavelet transform [DWT]
• CPC Class Codes: G06T 2207/30181 Earth observation
• CPC Class Codes: G06T 7/40 Analysis of texture depth or shape recovery from texture G06T7/529
• CPC Class Codes: G06V 10/446 using Haar-like filters, e.g. using integral image techniques

Abstract

Shift-invariant wavelet transform with properly selected wavelet base and decomposition level(s), is used to characterize rough-wavelet granules producing wavelet granulation of a feature space for a multispectral image such as a remote sensing image. Through the use of the granulated feature space contextual information in time and/or frequency domains are analyzed individually or in combination. Neighborhood rough sets (NRS) are employed in the selection of a subset of granulated features that further explore the local and/or contextual information from neighbor granules.

Claims

1. A method to perform rough-wavelet based analysis of spatio-temporal patterns, the method comprising:

generating a wavelet granulated space of features associated with a multispectral image, wherein the wavelet granulated space constitutes 4ⁿ granules in an n-dimension feature space for a one-level discrete wavelet transform (DWT) decomposition and 7ⁿ granules in the n-dimension feature space for a two-level DWT decomposition;

selecting features based on a rough set evaluation;

removing redundant features;

in response to removal of at least one of the redundant features, determining a measure of significance of the features by an evaluation of a change in data dependency of the features, wherein a greater change in data dependency indicates a greater measure of the significance;

classifying the spatio-temporal patterns based on the selected features; and

locating and selecting a subset of granulated features based on the significance.

2. The method according to claim 1, further comprising using spectral band values as the features.

3. The method according to claim 1, further comprising:

generating the wavelet granulated space by formulating shift-invariant granules in wavelet domain; and

formulating the shift-invariant granules through a shift-invariant discrete wavelet transform (SI-DWT) that characterizes feature values for wavelet granulation of an input pattern vector of the multispectral image.

4. The method according to claim 1, further comprising selecting wavelet bases and decomposition levels to characterize granules for producing wavelet granulation of the n-dimension feature space.

5. The method according to claim 1, wherein selecting the features based on the rough set evaluation comprises employing neighborhood rough sets (NRS) in the selection of the subset of granulated features that further explore at least one from a set of:

local and contextual information from neighbor granules.

6. The method according to claim 5, wherein classifying the patterns based on the selected features comprises determining a threshold for distances used in NRS, beyond which classification performance falls, and the method further comprises:

evaluating the subset of granulated features based on a significance of the features; and

employing a forward greed search algorithm for the feature selection.

7. The method according to claim 1, further comprising employing one from a set of:

a k-nearest neighbors classifier, a maximum likelihood classifier, and a multilayered perceptron classifier to classify the patterns based on the selected features.

8. The method according to claim 7, wherein the k-nearest neighbors classifier uses one from a set of:

k=1, k=3, and k=5.

9. An apparatus to perform rough-wavelet based analysis of spatio-temporal patterns, comprising:

a memory configured to store instructions and data associated with an input pattern vector of a multispectral image;

a processor coupled to the memory, wherein the processor is adapted to execute the instructions, which when executed configure the processor to;

generate a wavelet granulated space of features associated with the multispectral image, wherein the wavelet granulated space constitutes 4ⁿ granules in an n-dimension feature space for a one-level discrete wavelet transform (DWT) decomposition and 7ⁿ granules in the n-dimension feature space for a two-level DWT decomposition;

select features based on a rough set evaluation;

remove redundant features; and

classify the patterns based on the selected features.

10. The apparatus according to claim 9, wherein the rough-wavelet based analysis is performed to generate a model for land cover classification of a multispectral remote sensing image, and wherein the wavelet granulated space is generated by formulating shift-invariant granules in wavelet domain through a shift-invariant DWT (SI-DWT) that characterizes feature values for wavelet granulation of the input pattern vector of the multispectral remote sensing image.

11. The apparatus according to claim 10, wherein the processor is further configured to stop the SI-DWT after a second decomposition.

12. The apparatus according to claim 9, wherein rough-wavelet based analysis is performed for one from a set of:

texture detection and image indexing.

13. The apparatus according to claim 9, wherein the processor is further configured to select wavelet bases and decomposition levels to characterize granules to produce wavelet granulation of the n-dimension feature space.

14. The apparatus according to claim 13, wherein the processor is further configured to employ bi-orthogonal wavelet bases.

15. The apparatus according to claim 9, wherein the processor is further configured to employ neighborhood rough sets (NRS) in the selection of a subset of granulated features that further explore at least one from a set of:

local and contextual information from neighbor granules, and to evaluate the subset of granulated features based on a significance of the features.

16. A non-transitory computer-readable storage medium having instructions stored thereon to perform rough-wavelet based analysis of spatio-temporal patterns, the instructions comprising:

generating a wavelet granulated space of features associated with a multispectral image, wherein the wavelet granulated space constitutes 4ⁿ granules in an n-dimension feature space for a one-level discrete wavelet transform (DWT) decomposition and 7ⁿ granules in the n-dimension feature space for a two-level DWT decomposition;

selecting features based on a rough set evaluation;

removing redundant features; and

classifying the patterns based on the selected features.

17. The non-transitory computer-readable storage medium according to claim 16, wherein the instructions further comprise:

generating the wavelet granulated space is by formulating shift-invariant granules in wavelet domain through a shift-invariant DWT (SI-DWT) that characterizes feature values for wavelet granulation of an input pattern vector of the multispectral image; and

stopping the SI-DWT after a second decomposition.

18. The non-transitory computer-readable storage medium according to claim 16, wherein the instructions further comprise:

selecting wavelet bases and decomposition levels to characterize granules to produce wavelet granulation of the n-dimension feature space; and

employing bi-orthogonal wavelet bases.

19. The non-transitory computer-readable storage medium according to claim 16, wherein selecting the features based on the rough set evaluation comprises employing neighborhood rough sets (NRS) in the selection of a subset of granulated features that further explore at least one from a set of:

local and contextual information from neighbor granules and evaluating the subset of granulated features based on a significance of the features.