CN112784747B - Multi-scale eigen decomposition method for hyperspectral remote sensing image - Google Patents

Multi-scale eigen decomposition method for hyperspectral remote sensing image Download PDF

Info

Publication number
CN112784747B
CN112784747B CN202110086560.3A CN202110086560A CN112784747B CN 112784747 B CN112784747 B CN 112784747B CN 202110086560 A CN202110086560 A CN 202110086560A CN 112784747 B CN112784747 B CN 112784747B
Authority
CN
China
Prior art keywords
scale
matrix
hyperspectral
image
hyperspectral image
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202110086560.3A
Other languages
Chinese (zh)
Other versions
CN112784747A (en
Inventor
谷延锋
谢雯
金旭东
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Harbin Institute of Technology
Original Assignee
Harbin Institute of Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Harbin Institute of Technology filed Critical Harbin Institute of Technology
Priority to CN202110086560.3A priority Critical patent/CN112784747B/en
Publication of CN112784747A publication Critical patent/CN112784747A/en
Application granted granted Critical
Publication of CN112784747B publication Critical patent/CN112784747B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/10—Terrestrial scenes
    • G06V20/13—Satellite images
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
    • G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
    • G06V10/464—Salient features, e.g. scale invariant feature transforms [SIFT] using a plurality of salient features, e.g. bag-of-words [BoW] representations
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/10—Terrestrial scenes
    • G06V20/194—Terrestrial scenes using hyperspectral data, i.e. more or other wavelengths than RGB

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Multimedia (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Computation (AREA)
  • Evolutionary Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • Astronomy & Astrophysics (AREA)
  • Remote Sensing (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Investigating Or Analysing Materials By Optical Means (AREA)

Abstract

A multi-scale eigen decomposition method for a hyperspectral remote sensing image relates to the technical field of remote sensing image processing, and aims at the problem that the reflectivity component precision of the hyperspectral image obtained in the prior art is low, and comprises the following steps: the method comprises the following steps: acquiring a hyperspectral image, and obtaining a sparse graph matrix under N scales according to the hyperspectral image; step two: obtaining an eigen decomposition matrix of the multi-scale hyperspectral image according to the sparse image matrix under the N scales; step three: the hyperspectral image is utilized, and geometric average is carried out on the spectral dimension to obtain
Figure DDA0002910976300000011
Then, geometric mean is carried out on the space dimension to obtain
Figure DDA0002910976300000012
Step four: from the sum of eigen decomposition matrices
Figure DDA0002910976300000013
And obtaining the reflectivity component of the hyperspectral image. Compared with the prior art, the method has higher precision in the reflectivity component of the hyperspectral image obtained by combining multiple scales and integrating two strategies.

Description

Multi-scale eigen decomposition method for hyperspectral remote sensing image
Technical Field
The invention relates to the technical field of remote sensing image processing, in particular to a multi-scale eigen decomposition method for a hyperspectral remote sensing image.
Background
In recent years, remote sensing imaging technology is continuously developed, and the method has important application in the fields of land cover monitoring, urban planning and the like, and has huge future development potential. The original hyperspectral images have different reflectivities among the same ground objects due to other reasons such as geometric distribution difference of the surfaces of the objects, so that the problem of same object and different spectrums is caused, and a new problem is brought to the subsequent processing of the hyperspectral images. The purpose of intrinsic decomposition is to recover the reflectivity component from the original data and remove the influence caused by illumination or shadow, which is an important preprocessing step of the image processing technology, and the quality of intrinsic decomposition directly determines the performance of the subsequent image processing algorithm.
The existing intrinsic decomposition method mainly aims at RGB images and natural images, and the images have simple scenes, small scale change and uniform illumination, so that the complex problems of multi-scale difference and the like in space do not need to be considered. Aiming at the problems of complex and various scenes, large change of spatial scale difference, same-object and different-spectrum and the like of the hyperspectral remote sensing images, the invention designs the multi-scale combined hyperspectral image intrinsic decomposition method, integrates two strategies to eliminate the spectrum difference between the same objects, has better pertinence and has better performance compared with the existing intrinsic decomposition method.
Disclosure of Invention
The purpose of the invention is: aiming at the problem that the reflectivity component precision of a hyperspectral image obtained in the prior art is low, a multiscale eigen decomposition method of a hyperspectral remote sensing image is provided.
The technical scheme adopted by the invention to solve the technical problems is as follows:
the multi-scale eigen decomposition method of the hyperspectral remote sensing image comprises the following steps:
the method comprises the following steps: acquiring a hyperspectral image, and obtaining a sparse graph matrix under N scales according to the hyperspectral image;
step two: obtaining an eigen decomposition matrix of the multi-scale hyperspectral image according to the sparse image matrix under the N scales;
step three: the hyperspectral image is utilized, and geometric average is carried out on the spectral dimension to obtain
Figure BDA0002910976280000011
Then, geometric mean is made on the space dimension to obtain
Figure BDA0002910976280000012
Step four: from the sum of eigen decomposition matrices
Figure BDA0002910976280000013
And obtaining the reflectivity component of the hyperspectral image.
Further, the step of acquiring the sparse graph matrix in the step one is as follows: firstly, a set N of all pixels in a k multiplied by k window with each pixel i as the center of the pixel i in the hyperspectral image H is foundiWherein
Figure BDA00029109762800000214
i is 1,2, …, n, then linear correlation coefficient vectors of the pixel i and all pixels in the window are calculated by utilizing sparse graph coding, and finally a sparse graph matrix W belonging to R of the hyperspectral image H under the scale is formed by utilizing the linear correlation coefficient vectors of all the pixelsn×nWhere n is the total number of pixels and k × k represents the window size.
Further, the solution optimization criterion of the sparse graph matrix is as follows:
Figure BDA0002910976280000021
wherein
Figure BDA0002910976280000022
Composed of all pixel vectors and identity matrices within the window except for pixel IdIs an identity matrix of d x d,
Figure BDA0002910976280000023
in the form of a vector of coefficients,
each element W (i, j) and alpha of the sparse graph matrixiThe following relationships exist between the elements:
Figure BDA0002910976280000024
where i is 1,2, …, n, j is 1,2, …, n.
Further, the eigen decomposition matrix of the multi-scale hyperspectral image in the second step is expressed as:
Figure BDA0002910976280000025
Figure BDA0002910976280000026
wherein, W1，W2，W3···WNRespectively represent the sparse graph matrixes under N kinds of scale windows,
Figure BDA0002910976280000027
representing a sparse graph matrix obtained by linear weighting of the sparse graph matrix under N kinds of scale windows
Figure BDA0002910976280000028
σ1,σ2,σ3…σNCorresponding to the weight coefficient of the sparse graph matrix under different k scales,
Figure BDA0002910976280000029
Inis an n × n identity matrix.
Further, the
Figure BDA00029109762800000210
Expressed as:
Figure BDA00029109762800000211
wherein H represents a hyperspectral image.
Further, the
Figure BDA00029109762800000212
Expressed as:
Figure BDA00029109762800000213
further, the hyperspectral reflectivity component is expressed as:
Figure BDA0002910976280000031
further, in the first step, a sparse graph matrix under 4 scales is obtained according to the hyperspectral image.
Further, N is 4, and the window size at each scale is k1＝3，k2＝5，k3＝7，k4＝9。
The invention has the beneficial effects that:
aiming at the problems of same objects and different spectrums caused by the reasons that the unique scene of the hyperspectral remote sensing image is complex and various and the scale difference is large, sparse graph matrixes under multiple scales are generated by utilizing sparse graph codes, the similarity relation of the reflectivity in the large scene adapting to the hyperspectral image is constructed, then two strategies are integrated to eliminate the spectrum change between the same objects, each strategy has respective side points and different optimization effects, the strategy 1 eliminates the spectrum change caused by the geometric distribution of the surface of the object, and the strategy 2 eliminates the spectrum change caused by the illumination changing along with the spatial distribution. Therefore, the reflectivity component of the hyperspectral image obtained by combining the multi-scale strategy and integrating the two strategies has higher precision compared with the prior art.
Drawings
FIG. 1 is a flow chart;
FIG. 2 is an experimental raw image;
FIG. 3 is an experimental truth image;
FIG. 4 is a graph of the results of intrinsic decomposition;
FIG. 5 is a graph of the results of a single-scale eigen decomposition method.
Detailed Description
It should be noted that, in the present invention, the embodiments disclosed in the present application may be combined with each other without conflict.
The first embodiment is as follows: this embodiment is specifically described with reference to fig. 1, and specifically includes the following steps:
step 1: according to a given heightAnd calculating a sparse graph matrix under N scales by the spectral image. Sparse graph matrix W ∈ Rn×nInput hyperspectral image H ═ H is described1,H2,…,Hn]∈Rd×nThe similarity degree between the medium pixels and the pixels corresponds to different window sizes in different scales, and the sparse graph matrix in N scales is represented as WmAnd m is 1,2, …, N. Where n is the total number of pixels, d represents the spectral dimension of the hyperspectral image H, and the window size is represented by k × k.
Firstly, a set N of all pixels in a k multiplied by k window with each pixel i as the center of the pixel i in the hyperspectral image H is foundiWherein
Figure BDA0002910976280000032
i is 1,2, …, n. And calculating linear correlation coefficient vectors of the pixel i and all pixels in the window by using sparse graph coding, wherein the linear correlation coefficient vectors of all the pixels form a sparse graph matrix W of the hyperspectral image H under the scale. The specific optimization criteria are as follows:
Figure BDA0002910976280000041
wherein
Figure BDA0002910976280000042
From all pixel vectors within the window except for pixel i (pixel vector here meansk 2-1Vector of individual pixel intensity values
Figure BDA0002910976280000043
) And identity matrix composition, IdIs an identity matrix of d x d,
Figure BDA0002910976280000044
is a coefficient vector. Each element W (i, j) and alpha of the sparse graph matrixiThere are the following relationships between:
Figure BDA0002910976280000045
αiis a vector, the subscript j represents its element index number,
Figure BDA0002910976280000046
represents alphaiThe jth element of (1). On the basis of experiments of a large amount of remote sensing data, the invention sets the scale number N to be 4 and the scale size k1＝3，k2＝5，k3＝7，k4Since N is 4, σ is required in claim 4, since k is 4, k is a physical meaning of 9, (k1, k2, k3, k4 and k × k, but we do not need to know the value of k1,σ2,σ3,σ4Due to the value of
Figure BDA0002910976280000047
(m is 1,2, …, N), so k is required1,k2,k3,k4The value of (c). The number of subscripts is therefore related to the value of N. ) Calculating a sparse graph matrix W under four scales1，W2，W3，W4。
Step 2: and calculating an eigen decomposition matrix of the multi-scale hyperspectral image. Inputting a sparse graph matrix W under four scales1，W2，W3，W4Obtaining weighted sparse graph matrix by linear weighting
Figure BDA0002910976280000048
The linear relationship is as follows:
Figure BDA0002910976280000049
wherein sigma1,σ2,σ3,σ4The weight coefficients corresponding to the sparse graph matrix under different k scales are set as the weight coefficients according to experimental experience
Figure BDA00029109762800000410
. The eigen decomposition matrix is defined by the formulaThe following:
Figure BDA00029109762800000411
wherein InIs an n × n identity matrix.
And step 3: and eliminating the spectral change between the same ground objects according to the two proposed strategies to obtain the reflectivity component of the hyperspectral image.
The hyperspectral remote sensing image has a mechanism of interaction between illumination and the surface of an object, so that the hyperspectral image H and the environmental illumination E have the following relationship in a logarithmic form:
Figure BDA0002910976280000051
where ρ is the reflectivity component, 1nAnd 1dAll 1 vectors, β ═ β, of nx 1 and dx 1, respectively1,β2,…,βn]TIs a direction matrix.
3a) The method comprises the following steps Firstly, considering the influence of the spectral dimension, inputting a hyperspectral image H, performing geometric averaging on the spectral dimension, and removing spectral change caused by geometric distribution of the surface of an object to obtain the hyperspectral image
Figure BDA0002910976280000052
The calculation formula of (2) is as follows:
Figure BDA0002910976280000053
3b) the method comprises the following steps In that
Figure BDA0002910976280000054
On the basis of the method, the influence of the space dimension is considered, namely, geometric mean is carried out on the space dimension, and the spectral change caused by illumination changing along with the space distribution is eliminated to obtain the method
Figure BDA0002910976280000055
Specific calculation formula is as followsThe following:
Figure BDA0002910976280000056
wherein InIs an n × n identity matrix.
Input eigen decomposition matrices G and
Figure BDA0002910976280000057
calculating the reflectivity component rho of the hyperspectral image, wherein the whole operation process is carried out in a logarithmic domain, so that the reflectivity component of the image needs to be converted back to the original domain when being solved, and the calculation formula is as follows:
Figure BDA0002910976280000058
the experiment designed by the invention comprises the following steps:
the data used in the experiment is a set of hyperspectral images taken by an airborne ROSIS sensor, the size of the hyperspectral images is 610X 340X 103, the original images and the truth images are shown in FIG. 2 and FIG. 3, FIG. 4 is the eigen-decomposition result of the method of the invention, FIG. 5 is the eigen-decomposition result under the single scale, and Table 1 is the classification accuracy comparison of the decomposition results of the two methods. The comparison result shows that the method provided by the invention has higher precision.
TABLE 1
Figure BDA0002910976280000059
It should be noted that the detailed description is only for explaining and explaining the technical solution of the present invention, and the scope of protection of the claims is not limited thereby. It is intended that all such modifications and variations be included within the scope of the invention as defined in the following claims and the description.

Claims (5)

1. The multi-scale eigen decomposition method of the hyperspectral remote sensing image is characterized by comprising the following steps of:
the method comprises the following steps: acquiring a hyperspectral image, and obtaining a sparse graph matrix under N scales according to the hyperspectral image;
step two: obtaining an eigen decomposition matrix of the multi-scale hyperspectral image according to the sparse image matrix under the N scales;
step three: the hyperspectral image is utilized, and geometric average is carried out on the spectral dimension to obtain
Figure FDA0003440335800000011
Then, geometric mean is carried out on the space dimension to obtain
Figure FDA0003440335800000012
Step four: from the sum of eigen decomposition matrices
Figure FDA0003440335800000013
Obtaining a high-spectrum reflectivity component;
the sparse graph matrix in the first step is obtained by the following steps: firstly, a set N of all pixels in a k multiplied by k window with each pixel i as the center of the pixel i in the hyperspectral image H is foundiWherein
Figure FDA0003440335800000014
Then calculating linear correlation coefficient vectors of the pixel i and all pixels in the window by utilizing sparse graph coding, and finally forming a sparse graph matrix W e R of the hyperspectral image H under the scale by utilizing the linear correlation coefficient vectors of all the pixelsn×nWhere n is the total number of pixels and k × k represents the window size;
the above-mentioned
Figure FDA0003440335800000015
Expressed as:
Figure FDA0003440335800000016
wherein H represents a hyperspectral image;
the above-mentioned
Figure FDA0003440335800000017
Expressed as:
Figure FDA0003440335800000018
the reflectance component of the hyperspectrum is expressed as:
Figure FDA0003440335800000019
wherein, IdIs a unit matrix of d × d, 1dIs a full 1 vector of dX 1, InIs an n × n identity matrix, 1nIs the full 1 vector of n x 1, and G is the eigen decomposition matrix of the multi-scale hyperspectral image.
2. The multi-scale eigen decomposition method of the hyperspectral remote sensing image according to claim 1, characterized in that the solving optimization criterion of the sparse graph matrix is:
Figure FDA00034403358000000110
wherein
Figure FDA00034403358000000111
Composed of all pixel vectors and identity matrices within the window except for pixel IdIs an identity matrix of d x d,
Figure FDA00034403358000000112
in the form of a vector of coefficients,
each element W (i, j) and alpha of the sparse graph matrixiThe following relationships exist between the elements:
Figure FDA0003440335800000021
where i is 1,2, …, n, j is 1,2, …, n.
3. The multi-scale eigen decomposition method for the hyperspectral remote sensing image according to claim 2, wherein the eigen decomposition matrix of the multi-scale hyperspectral image in the second step is represented as:
Figure FDA0003440335800000022
Figure FDA0003440335800000023
wherein, W1，W2，W3···WNRespectively represent the sparse graph matrixes under N kinds of scale windows,
Figure FDA0003440335800000024
representing a sparse graph matrix obtained by linear weighting of the sparse graph matrix under N kinds of scale windows
Figure FDA0003440335800000025
σ1,σ2,σ3…σNCorresponding to the weight coefficient of the sparse graph matrix under different k scales,
Figure FDA0003440335800000026
Inis an n × n identity matrix.
4. The multi-scale eigen decomposition method of the hyperspectral remote sensing image according to claim 1, characterized in that in the first step, a sparse map matrix under 4 scales is obtained according to the hyperspectral image.
5. The multi-scale eigen decomposition of the hyperspectral remote sensing image of claim 1The method is characterized in that the N is 4, and the window size under each scale is k1＝3，k2＝5，k3＝7，k4＝9。
CN202110086560.3A 2021-01-22 2021-01-22 Multi-scale eigen decomposition method for hyperspectral remote sensing image Active CN112784747B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110086560.3A CN112784747B (en) 2021-01-22 2021-01-22 Multi-scale eigen decomposition method for hyperspectral remote sensing image

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110086560.3A CN112784747B (en) 2021-01-22 2021-01-22 Multi-scale eigen decomposition method for hyperspectral remote sensing image

Publications (2)

Publication Number Publication Date
CN112784747A CN112784747A (en) 2021-05-11
CN112784747B true CN112784747B (en) 2022-03-04

Family

ID=75758520

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110086560.3A Active CN112784747B (en) 2021-01-22 2021-01-22 Multi-scale eigen decomposition method for hyperspectral remote sensing image

Country Status (1)

Country Link
CN (1) CN112784747B (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113408635B (en) * 2021-06-29 2022-03-08 哈尔滨工业大学 Hyperspectral image eigen decomposition method based on assistance of digital surface model
CN117197137B (en) * 2023-11-07 2024-02-09 山东大学 Tissue sample analysis method and system based on hyperspectral image

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104484884A (en) * 2014-12-30 2015-04-01 天津大学 Intrinsic image decomposition method based on multi-scale L0 sparse constraint
CN107316309A (en) * 2017-06-29 2017-11-03 西北工业大学 High spectrum image conspicuousness object detection method based on matrix decomposition
CN111899257A (en) * 2020-08-14 2020-11-06 哈尔滨工业大学 Ground object spectral reflectivity image extraction method based on multi-temporal intrinsic image decomposition

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104484884A (en) * 2014-12-30 2015-04-01 天津大学 Intrinsic image decomposition method based on multi-scale L0 sparse constraint
CN107316309A (en) * 2017-06-29 2017-11-03 西北工业大学 High spectrum image conspicuousness object detection method based on matrix decomposition
CN111899257A (en) * 2020-08-14 2020-11-06 哈尔滨工业大学 Ground object spectral reflectivity image extraction method based on multi-temporal intrinsic image decomposition

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
Intrinsic Image Recovery From Remote;Xudong Jin等;《IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING》;20190131;全文 *

Also Published As

Publication number Publication date
CN112784747A (en) 2021-05-11

Similar Documents

Publication Publication Date Title
Qian et al. Hyperspectral imagery restoration using nonlocal spectral-spatial structured sparse representation with noise estimation
CN107316309B (en) Hyperspectral image saliency target detection method based on matrix decomposition
CN108416723B (en) Lens-free imaging fast reconstruction method based on total variation regularization and variable splitting
CN108537788B (en) Camouflage effect evaluation method and device, computer equipment and storage medium
CN112784747B (en) Multi-scale eigen decomposition method for hyperspectral remote sensing image
CN102542542A (en) Image denoising method based on non-local sparse model
CN107798345B (en) High-spectrum disguised target detection method based on block diagonal and low-rank representation
CN110490894B (en) Video foreground and background separation method based on improved low-rank sparse decomposition
CN111680579B (en) Remote sensing image classification method for self-adaptive weight multi-view measurement learning
CN112633202B (en) Hyperspectral image classification algorithm based on dual denoising combined multi-scale superpixel dimension reduction
CN113033602B (en) Image clustering method based on tensor low-rank sparse representation
Aghamaleki et al. Image fusion using dual tree discrete wavelet transform and weights optimization
CN109858531B (en) Hyperspectral remote sensing image fast clustering algorithm based on graph
CN113421198B (en) Hyperspectral image denoising method based on subspace non-local low-rank tensor decomposition
Shahdoosti et al. A new compressive sensing based image denoising method using block-matching and sparse representations over learned dictionaries
Xu et al. AACNet: Asymmetric attention convolution network for hyperspectral image dehazing
CN110880192B (en) Image DCT coefficient distribution fitting method based on probability density function dictionary
Wang et al. Adaptive feature fusion network based on boosted attention mechanism for single image dehazing
CN110009575B (en) Infrared image stripe noise suppression method based on sparse representation
CN110717485A (en) Hyperspectral image sparse representation classification method based on local preserving projection
CN111062888A (en) Hyperspectral image denoising method based on multi-target low-rank sparsity and spatial-spectral total variation
Yufeng et al. Research on SAR image change detection algorithm based on hybrid genetic FCM and image registration
CN112785583B (en) Hyperspectral remote sensing image reflectivity recovery method based on superpixel segmentation
CN110378268A (en) EO-1 hyperion fast target detection method and system
CN112785662B (en) Self-adaptive coding method based on low-resolution first-pass information

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant