Method and system for extracting and classifying features of hyperspectral remote sensing image

Patent metadata

Field	Value
Publication number	US-10509984-B2
Application number	US-201815978189-A
Country	US
Kind code	B2
Filing date	May 13, 2018
Priority date	Nov 1, 2016
Publication date	Dec 17, 2019
Grant date	Dec 17, 2019

Abstract

The present invention provides a method for extracting and classifying features of hyperspectral remote sensing image, including: an sampling step, a binarizing step, a coding step, a statistical calculating step, a concatenating step, and a classifying step. The present invention further provides a system for extracting and classifying features of hyperspectral remote sensing image. The technical solution provided by the present invention can make full use of the contextual relationship between the spectral domain and the spatial domain in a hyperspectral remote sensing image by extending two-dimensional LBPs into three-dimensional LBPs, and has good robustness to noise by introducing a relaxation threshold discrimination operation. Furthermore, the rotation-invariant three-dimensional LBP model provided by the present invention takes account of the essential characteristics of the hyperspectral remote sensing image, and therefore the present solution has advantages that it is targeted, simple in operation and high in calculation efficiency.

First claim

What is claimed is:

1. A method for extracting and classifying features of a hyperspectral remote sensing image, the method comprising:
a step of sampling, in which, for each respective pixel in the hyperspectral remote sensing image, six vertex pixels in an octahedral domain of a central pixel corresponding to the each respective pixel are used as sampling points in a three-dimensional Local Binary Pattern (3DLBP), and a local spatial-spectral structure of the central pixel is described by using a gray level distribution T of the six sampling points, where T≈t(g 0 , g 1 , g 2 , g 3 , g 4 , g 5 ), which represents the gray level distribution of the six sampling points, with g 0 , g 1 , g 2 , g 3 , g 4 , and g 5 representing a pixel value of the six sampling points respectively;
a step of binarizing, in which gray values of the six vertex pixels in the octahedral domain are respectively compared with a pixel value of the central pixel, and if an absolute value of a difference between values of the central pixel and a corresponding vertex pixel among the six vertex pixels is lower than a predefined discriminant threshold, label the corresponding vertex pixel as 1, otherwise, label the corresponding vertex pixel as 0, so as to form a binary pattern T1 for the local spatial-spectral structure of the central pixel, where T1≈t(s(g 0 −g c , s(g 1 −g c ), . . . , s(g 5 −g c )), with g c representing the pixel value of the central pixel and s representing a binary function;
a step of coding, in which each of the binary patterns having a same spatial topology is uniquely labeled with the number of 1s in the binary pattern, so as to obtain a 3DLBP code for each of the binary patterns, that is,

3
⁢
DLBP

=

{

Σ

p
=
0

5

⁢

s
⁡

(

g
p

-

g
c

)

,

if
⁢

⁢

Γ
⁡

(

T
⁢

⁢
1

)

≤
6

7
⁢

,
else

⁢

,

where the spatial topology of the binary pattern is calculated by equation:

Γ
⁡

(

T
⁢

⁢
1

)

=

∑

p
=
0

5

⁢

⁢

|

s
⁡

(

g
i

-

g
c

)

-

s
⁡

(

g
j

-

g
c

)

|

,

such that binary patterns with the same value of Γ are determined to have the same spatial topology, in which g i and g j represent the ith and jth sampling points in the binary pattern respectively, g i and g j are adjacent to each other, and i≠j, i, j∈{0, 1, 2, 3, 4, 5};
a step of statistical calculating, in which after the 3DLBP code is obtained for each pixel, a co-occurrence frequency for eight 3DLBP codes of 0, 1, 2, 3, 4, 5, 6, and 7 is statistically calculated within an n×n rectangular neighborhood of each of the pixels in an image element, so as to obtain a histogram feature of the pixel;
a step of concatenating, in which the histogram features of all the pixels in the image element are concatenated sequentially, so as to obtain a 3DLBP feature c

Assignees

• Univ Shenzhen

Inventors

• Jia Sen
• Hu Jie
• Deng Lin

Classifications

• G06V20/13 Primary

  Satellite images · CPC title

• G06V10/50

  by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis · CPC title

• G06F18/24

  Classification techniques · CPC title

• G06K9/6212 Primary

  Physics · mapped topic

• G06K9/46

  Physics · mapped topic