CN108734672B - Hyperspectral data unmixing method based on spectral library cutting and collaborative sparse regression - Google Patents
Hyperspectral data unmixing method based on spectral library cutting and collaborative sparse regression Download PDFInfo
- Publication number
- CN108734672B CN108734672B CN201810016715.4A CN201810016715A CN108734672B CN 108734672 B CN108734672 B CN 108734672B CN 201810016715 A CN201810016715 A CN 201810016715A CN 108734672 B CN108734672 B CN 108734672B
- Authority
- CN
- China
- Prior art keywords
- library
- spectrum
- spectral
- representing
- spectral library
- 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
Links
- 230000003595 spectral effect Effects 0.000 title claims abstract description 118
- 238000000034 method Methods 0.000 title claims abstract description 57
- 238000001228 spectrum Methods 0.000 claims abstract description 89
- 239000011159 matrix material Substances 0.000 claims abstract description 62
- 229910052500 inorganic mineral Inorganic materials 0.000 claims abstract description 29
- 239000011707 mineral Substances 0.000 claims abstract description 29
- OAICVXFJPJFONN-UHFFFAOYSA-N Phosphorus Chemical compound [P] OAICVXFJPJFONN-UHFFFAOYSA-N 0.000 claims description 9
- 230000003190 augmentative effect Effects 0.000 claims description 4
- 238000012544 monitoring process Methods 0.000 abstract description 3
- 230000007613 environmental effect Effects 0.000 abstract description 2
- 238000005457 optimization Methods 0.000 abstract 1
- BERDEBHAJNAUOM-UHFFFAOYSA-N copper(I) oxide Inorganic materials [Cu]O[Cu] BERDEBHAJNAUOM-UHFFFAOYSA-N 0.000 description 21
- LBJNMUFDOHXDFG-UHFFFAOYSA-N copper;hydrate Chemical compound O.[Cu].[Cu] LBJNMUFDOHXDFG-UHFFFAOYSA-N 0.000 description 21
- 229910052934 alunite Inorganic materials 0.000 description 10
- 239000010424 alunite Substances 0.000 description 10
- KPZTWMNLAFDTGF-UHFFFAOYSA-D trialuminum;potassium;hexahydroxide;disulfate Chemical compound [OH-].[OH-].[OH-].[OH-].[OH-].[OH-].[Al+3].[Al+3].[Al+3].[K+].[O-]S([O-])(=O)=O.[O-]S([O-])(=O)=O KPZTWMNLAFDTGF-UHFFFAOYSA-D 0.000 description 10
- VYPSYNLAJGMNEJ-UHFFFAOYSA-N Silicium dioxide Chemical compound O=[Si]=O VYPSYNLAJGMNEJ-UHFFFAOYSA-N 0.000 description 9
- 239000011045 chalcedony Substances 0.000 description 9
- 239000010433 feldspar Substances 0.000 description 8
- 230000000694 effects Effects 0.000 description 4
- 238000010521 absorption reaction Methods 0.000 description 3
- 239000000126 substance Substances 0.000 description 3
- 102000008186 Collagen Human genes 0.000 description 2
- 108010035532 Collagen Proteins 0.000 description 2
- 229920001436 collagen Polymers 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 238000004088 simulation Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000012880 independent component analysis Methods 0.000 description 1
- 238000003331 infrared imaging Methods 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/50—Image enhancement or restoration using two or more images, e.g. averaging or subtraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10032—Satellite or aerial image; Remote sensing
- G06T2207/10036—Multispectral image; Hyperspectral image
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A40/00—Adaptation technologies in agriculture, forestry, livestock or agroalimentary production
- Y02A40/10—Adaptation technologies in agriculture, forestry, livestock or agroalimentary production in agriculture
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Investigating Or Analysing Materials By Optical Means (AREA)
Abstract
The invention discloses a hyperspectral data unmixing method based on spectral library cutting and collaborative sparse regression, which comprises the following steps of: inputting a spectrum library and hyperspectral data to be unmixed; initializing a spectrum library and constructing a spectrum data matrix; constructing a spectrum fitting error fidelity term; constructing an abundance matrix cooperative sparsity constraint term; constructing a sparse regularization item cut by a spectrum library; establishing a spectral library cutting and collaborative sparse regression model; iteratively solving a spectral library cutting and collaborative sparse regression model; and outputting the cut spectral library and the end-member abundance map. In order to relieve the problem of mismatching between end members of a spectrum library and end members in an actual scene, a collaborative sparse regression model is established, objective function optimization unmixing is carried out, the accuracy of unmixing is improved, the mismatching rate of the end members is reduced, and the robustness to spectral amplitude variation and noise is enhanced; the method can be widely applied to the application of high-spectrum data unmixing in the fields of environmental monitoring, mineral exploration, precision agriculture and the like.
Description
Technical Field
The invention relates to a remote sensing hyperspectral data processing technology, in particular to a hyperspectral data unmixing method based on spectral library cutting and collaborative sparse regression.
Background
The hyperspectral data is widely applied to the fields of military monitoring, precise agriculture, mineral exploration and the like due to spectral correlation and abundant spatial information. The hyperspectral data unmixing is a key technology of quantitative remote sensing analysis. The basic principle of hyperspectral data unmixing is to decompose a single pixel spectrum into a combination of a plurality of pure pixel spectra. The theoretical basis is that due to the limitation of the spatial resolution of an imaging spectrometer, a large number of mixed pixels exist in an obtained hyperspectral image, and each mixed pixel contains multiple pure substances.
Many unmixing algorithms for hyperspectral data have been proposed at present, including pure pixel index, vertex component analysis, independent component analysis, and related component analysis. However, the methods in the above categories are all to extract end members from hyperspectral data. Through research for decades, people have collected standard spectra of a plurality of minerals to form a huge spectral library, and the spectral library can be completely utilized to directly perform spectral unmixing without the step of end member extraction.
2014 Li Yun Song et al proposed a hyperspectral image sparse unmixing method based on neighborhood spectral weighting [ Sigan electronics university, China, hyperspectral image sparse unmixing method based on neighborhood spectral weighting [ p ] application for invention, CN103810715A,2014-03-12], and obtained a good unmixing effect. However, due to the influence of temperature, illumination and the like, the spectral curve of the same mineral collected in a laboratory and the spectral curve collected in a real environment have deviation. The method can not effectively process the spectrum change phenomenon, and the unmixing effect of the algorithm is reduced when a large amount of noise exists in the data.
Disclosure of Invention
The invention aims to provide a hyperspectral data unmixing method based on spectral library cutting and collaborative sparse regression.
The technical solution for realizing the purpose of the invention is as follows: a hyperspectral data unmixing method based on spectral library cutting and collaborative sparse regression comprises the following steps:
step 1, inputting a spectrum library and hyperspectral data to be unmixed;
step 2, initializing a spectrum library and constructing a spectrum data matrix;
step 3, constructing a spectrum fitting error fidelity term;
step 4, constructing an abundance matrix cooperative sparsity constraint term;
step 5, constructing a cut sparse regularization item of the spectrum library;
step 6, establishing a spectral library cutting and collaborative sparse regression model;
step 7, iteratively solving a spectral library cutting and collaborative sparse regression model;
and 8, outputting the cut spectral library and the end-member abundance map.
Compared with the prior art, the invention has the remarkable advantages that: (1) the spectrum mismatching phenomenon existing in the sparse unmixing process is fully considered, the priori knowledge with sparsity in spectrum change is utilized, the collaborative sparse regression framework is combined, the spectrum mismatching phenomenon is modeled, compared with the traditional sparse unmixing method, the sparse unmixing method has the advantages that the unmixing effect is improved, and the robustness to spectrum amplitude change and noise is enhanced; (2) the method can be widely applied to the application of high-spectrum data unmixing in the fields of environmental monitoring, mineral exploration, precision agriculture and the like.
The present invention is described in further detail below with reference to the attached drawing figures.
Drawings
FIG. 1 is a flow chart of a hyperspectral data unmixing method based on spectral library clipping and collaborative sparse regression.
Figure 2(a) is an alunite mineral abundance map obtained using the tecorder software on the Cuprite dataset.
Fig. 2(b) is an alunite mineral abundance map obtained using the SUnSAL method on the Cuprite dataset.
Fig. 2(c) is an alunite mineral abundance map obtained using the CLSUnSAL method on the Cuprite dataset.
Fig. 2(d) is an alunite mineral abundance map obtained using the DANSER method on the Cuprite dataset.
Fig. 2(e) is an alunite mineral abundance map obtained using the SDCSR method on the Cuprite dataset.
Figure 2(f) is a plot of hydroammonium feldspar mineral abundance on the Cuprite dataset using the tecorder software.
Fig. 2(g) is a plot of the abundance of hydroammonium feldspar minerals on the Cuprite dataset using the SUnSAL method.
Fig. 2(h) is a plot of the abundance of hydroammonium feldspar minerals on the Cuprite dataset using the CLSUnSAL method.
Figure 2(i) is a plot of the abundance of hydroammonium feldspar minerals on the Cuprite dataset using the DANSER method.
Figure 2(j) is a plot of the abundance of hydroammonium feldspar minerals on the Cuprite dataset using the SDCSR method.
Fig. 2(k) is a map of chalcedony mineral abundance on the Cuprite dataset using the tecorder software.
Fig. 2(l) is a plot of chalcedony mineral abundance on the Cuprite dataset using the SUnSAL method.
Fig. 2(m) is a plot of chalcedony mineral abundance on the Cuprite dataset using the CLSUnSAL method.
Fig. 2(n) is a plot of chalcedony mineral abundance on the Cuprite dataset using the DANSER method.
Fig. 2(o) is a plot of chalcedony mineral abundance on the Cuprite dataset using the SDCSR method.
FIG. 3(a) is a graph comparing the alunite spectra curves in the initial spectral library and the spectral library after clipping.
FIG. 3(b) is a comparison of the spectrum curves of the hydroammonium feldspar in the initial spectrum library and the spectrum library after cutting.
FIG. 3(c) is a comparison of the initial spectral library and the clipped spectral library for the chalcedony spectral curves.
Detailed Description
With reference to fig. 1, a hyperspectral data unmixing method based on spectral library clipping and collaborative sparse regression includes the following steps:
step 1, inputting a spectrum library and hyperspectral data to be unmixed;
step 2, initializing a spectrum library and constructing a spectrum data matrix;
step 3, constructing a spectrum fitting error fidelity term;
step 4, constructing an abundance matrix cooperative sparsity constraint term;
step 5, constructing a cut sparse regularization item of the spectrum library;
step 6, establishing a spectral library cutting and collaborative sparse regression model;
step 7, iteratively solving a spectral library cutting and collaborative sparse regression model;
and 8, outputting the cut spectral library and the end-member abundance map.
Further, the specific process of inputting the spectrum library and the hyperspectral data to be unmixed in the step 1 is as follows:
inputting a spectrum library containing hyperspectral end members
Wherein the integer K>0 denotes the number of bands per spectral signal, integer N>0 represents the number of end-member spectral signals contained in the spectral library;
inputting hyperspectral data to be unmixed
Wherein the integer L>0 represents the number of bands of the hyperspectral data, an integer W>0 and an integer H>0 represents the width and height, respectively, of the hyperspectral data space dimension.
Further, step 2, the initialization of the spectrum library and the construction of the spectrum data matrix specifically comprise the following steps:
step 2-1, inputting a spectrum library end member
The number of wave bands is K>0 in a band of
And inputting a hyperspectral signal to be unmixed
Number of wave bands L>0 Band ═ b1,b2,...,bL]. According to the hyperspectral signal to be unmixed
Band of (1) and reserve end members of the spectrum library
Wave band
The other corresponding wave bands are discarded; through the one-step operation, the end member signals in the spectral library
The wave band of (1) and the hyperspectral signal to be unmixed
The wave bands in the spectrum library are kept consistent to obtain the end members of the spectrum library after the wave bands are removed
Wave bandNumber L>0 Band ═ b1,b2,...,bL]；
Step 2-2, inputting a spectrum library of
The number of wave bands is K>0, the number of the included end members is N>0, for spectral library
The ith end-member signal of
I is more than or equal to 1 and less than or equal to N, the wave band is removed by using the method in the step 2-1, and the end member signal after removal is obtained
Spectral library
By applying the method to each end member signal in the spectrum library, the spectrum library with the wave bands removed can be obtained
Step 2-3, initially inputting the hyperspectral data to be unmixed into
The hyperspectral data to be unmixed
Arranged pixel by pixel to form a spectral data matrix
Wherein the integer L>0 represents the number of wave bands, M is W multiplied by H represents the number of hyperspectral pixels,
representing a hyperspectral pixel with a spectral data matrix Y as a modeOne input of the type.
Further, a fitting error fidelity term between the hyperspectral data to be unmixed and the spectrum library matrix is constructed in the step 3:
in the formula
Representing a hyperspectral data matrix to be unmixed,
representing the spectral library to be tailored,
is an initialized abundance coefficient matrix, where L>0 denotes the number of bands, M>0 represents the number of spectral signals in the hyperspectral data matrix, N>0 represents the number of end member spectra contained in the spectral library.
Further, the construction of the abundance matrix collaborative sparsity constraint term in the step 4 specifically includes:
in the formula
Representing a matrix of abundance coefficients, | · | | non-conducting phosphor2,1Is represented by2,1Norm, Ui,jRepresenting a hyperspectral data matrix to be unmixed
Spectrum library corresponding to jth pixel
The abundance coefficient of the ith end member is that j is more than or equal to 1 and less than or equal to M, and i is more than or equal to 1 and less than or equal to N.
Further, the constructing of the tailored sparse regularization term of the spectrum library in the step 5 specifically comprises:
in the formula
Representing the spectral library after initialization,
representing the spectral library to be tailored,
to represent
The value corresponding to the p-th row and the q-th column in (1),
represents the value corresponding to the p-th row and the q-th column in D, p is more than or equal to 1 and less than or equal to L, q is more than or equal to 1 and less than or equal to N, | | · | | survival1,1Is represented by1,1And (4) norm.
Further, the specific steps of establishing the spectral library cutting and collaborative sparse regression model in the step 6 are as follows:
in the formula, the parameter lambda>0,β>0，
Representing a hyperspectral data matrix to be unmixed,
representing the spectral library after initialization,
representing the spectral library to be tailored,
representing a matrix of abundance coefficients, | · | | non-conducting phosphorFRepresenting the F norm of the matrix, | · | | non-woven phosphor2,1Is represented by2,1Norm, | · | luminance1,1Is represented by1,1And (4) norm.
Further, the iterative spectral library clipping and collaborative sparse regression model in step 7 specifically comprises the following steps:
(1) firstly, inputting an initialized spectrum library
And the hyperspectral data matrix Y to be unmixed is [ Y ═ Y1,y2,...,yM]∈RL×M；
(2) Initializing an abundance coefficient matrix
And a spectral library to be tailored
(3) Introducing a relaxation variable
The spectral library cutting and collaborative sparse regression model is changed into the following model:
parameter α in the formula>0,λ>0,β>0, introducing an equality constraint U ═ V to the model1,U＝V2Then solving the augmented Lagrange function minimization model:
in the formula, the parameter alpha>0,λ>0,β>0,μ>0, variable
The abundance coefficient matrix can be obtained after the solution by using the exchange direction multiplier method
Comprises the following steps:
wherein U is(k+1)Represents the abundance matrix U, V in the k +1 iteration1 (k)Representing the variable V during the kth iteration1，D1 (k)Representing the Lagrange multiplier variable D during the kth iteration1，V2 (k)Representing the variable V during the kth iteration2，D2 (k)Representing the Lagrange multiplier variable D during the kth iteration2To find out the spectrum library to be cut
Comprises the following steps:
wherein,
representing the spectrum library to be cut in the k +1 iteration process
Column i, D(:,i)Indicates the ith column, H in the initialization spectral library D(:,i) (k+1)The ith column, representing the relaxation variable H during the (k + 1) th iteration, is the soft threshold shrinkage function.
Further, the specific process of outputting the tailored spectrum library and the end-member abundance map in step 8 is as follows:
(1) output cut spectral library
(2) Matrix abundance coefficient
Restructuring into three-dimensional data
Then output
Wherein H represents the height of the mineral abundance map, W represents the width of the mineral abundance map, and N represents the spectral library after cutting
The number of end members in (1).
The present invention will be described in detail with reference to specific examples.
Examples
With reference to fig. 1, a hyperspectral data unmixing method based on spectral library clipping and collaborative sparse regression includes the following steps:
step 1, inputting a spectrum library and hyperspectral data to be unmixed: inputting a spectrum library containing hyperspectral end members
A spectral library refers to a collection of spectral signals containing a number of pure substances (also called end-members), where K represents the number of bands of each spectral signal and N represents the number of end-member spectral signals contained in the spectral library. The USGS spectral library is used as the input spectral library in this embodiment. In the USGS spectral library, K is 224 and N is 342. Inputting hyperspectral data to be unmixed
Wherein L represents the wave band number of the hyperspectral data, and W and H respectively represent the hyperspectral data spaceWidth and height of the dimension. Taking the Cuprite dataset as the hyperspectral data to be unmixed, wherein L is 188, W is 191, and H is 250 in the Cuprite dataset.
Step 2, initializing a spectrum library and constructing a spectrum data matrix:
(1) inputting a spectral library end member
The number of wave bands is K>0 in a band of
And inputting a hyperspectral signal to be unmixed
Number of wave bands L>0 Band ═ b1,b2,...,bL]. According to the hyperspectral signal to be unmixed
Band of (1) and reserve end members of the spectrum library
Wave band
The other bands are discarded. The end-member signal waveband range in the USGS spectrum is 1-224, and the total number of the end-member signal waveband ranges from 224 wavebands, while the total number of the end-member signal waveband ranges from 118 wavebands in the Cuprice data set. In the Cuprivate data set, the values of the wave bands such as 1-2, 105-115, 150-170, 223-224 and the like are removed due to the reasons of low water absorption and signal-to-noise ratio and the like, so that the end-member wave band of the USGS spectrum library is eliminated
The corresponding 1-2, 105-115, 150-170, 223-224 bands are also removed. Through the one-step operation, the wave band of the end member signal in the spectrum library USGS is consistent with the wave band in the hyperspectral data Cuprice.
(2) The input spectrum library is
The number of wave bands is K>0, the number of the included end members is N>0, for spectral library
The ith end-member signal of
I is more than or equal to 1 and less than or equal to N, the wave band is removed by using the method in (1), and the end member signal after removal is obtained
Spectral library
By applying the method, the spectrum library with the removed wave bands (namely initialized) can be obtained
And obtaining the USGS spectrum library after the wave bands are removed through the step.
(3) Initially input hyperspectral data to be unmixed into
Wherein the integer L>0 represents the number of wave bands of the hyperspectral data to be unmixed, and the integer W>0 and an integer H>0 represents the width and height of the spatial dimension of the hyperspectral data to be unmixed, respectively. The hyperspectral data to be unmixed
Arranged pixel by pixel to form a spectral data matrix
Wherein the integer L>0 represents the number of wave bands, M is W multiplied by H represents the number of hyperspectral pixels,
and representing a hyperspectral pixel, and taking a spectrum data matrix Y as an input of the model. Cuprite dataset
In the above formula, L188, W191, and H250 are used to arrange the Cuprite data sets pixel by pixel to obtain a spectral data matrix
L＝188,M＝191×250＝47750。
Step 3, constructing a spectrum fitting error fidelity term:
in the formula
Representing a hyperspectral data matrix to be unmixed,
representing the spectral library to be tailored,
is an initialized abundance coefficient matrix, where L>0 denotes the number of bands, M>0 represents the number of spectral signals in the hyperspectral data matrix, N>0 represents the number of end member spectra contained in the spectral library.
Step 4, constructing an abundance matrix collaborative sparsity constraint term:
in the formula
Representing a matrix of abundance coefficients, | · | | non-conducting phosphor2,1Is represented by2,1Norm, Ui,jRepresenting a hyperspectral data matrix to be unmixed
The spectral library corresponding to the jth pixel
The abundance coefficient of the ith end member is that j is more than or equal to 1 and less than or equal to M, i is more than or equal to 1 and less than or equal to N, and N>0 denotes the number of end-member spectra in the spectral library, M>And 0 represents the number of pixels in the hyperspectral data matrix.
Step 5, establishing a sparse regularization item for spectral library cutting:
in the formula
Representing the spectral library after initialization,
representing the spectral library to be tailored,
to represent
The value corresponding to the p-th row and the q-th column in (1),
represents the value corresponding to the p-th row and the q-th column in D, p is more than or equal to 1 and less than or equal to L, q is more than or equal to 1 and less than or equal to N, | | · | | survival1,1Is represented by1,1And (4) norm. Wherein L is>0 denotes the number of bands per spectral signal, N>0 represents the number of end-member spectra in the spectral library.
Step 6, solving a spectral library cutting and collaborative sparse regression model:
in the formula, the parameter lambda>0,β>0，
Representing a hyperspectral data matrix to be unmixed,
representing the spectral library after initialization,
representing the spectral library to be tailored,
representing a matrix of abundance coefficients, | · | | non-conducting phosphorFRepresenting the F norm of the matrix, | · | | non-woven phosphor2,1Is represented by2,1Norm, | · | luminance1,1Is represented by1,1And (4) norm.
Step 7, solving a spectral library cutting and collaborative sparse regression model:
(1) firstly, inputting an initialized spectrum library
And the hyperspectral data matrix Y to be unmixed is [ Y ═ Y1,y2,...,yM]∈RL×MWherein L is>0 denotes the number of bands per spectral signal, N>0 denotes the number of end-member spectra in the spectral library, M>And 0 represents the number of pixels in the hyperspectral data matrix to be unmixed.
(2) Initializing an abundance coefficient matrix
And a spectral library to be tailored
(3) Introducing a relaxation variable
Cutting and cooperating the spectrum libraryThe sparse regression model becomes:
parameter α in the formula>0,λ>0,β>0. Introducing equality constraint U-V into the model1,U＝V2Then solving the augmented Lagrange function minimization model:
in the formula, the parameter alpha>0,λ>0,β>0,μ>0, variable
The abundance coefficient matrix can be obtained after the solution by using the exchange direction multiplier method
Comprises the following steps:
wherein U is(k+1)Represents the abundance matrix U, V in the k +1 iteration1 (k)Representing the variable V during the kth iteration1，D1 (k)Representing the Lagrange multiplier variable D during the kth iteration1，V2 (k)Representing the variable V during the kth iteration2，D2 (k)Representing the Lagrange multiplier variable D during the kth iteration2. Obtaining a spectrum library to be cut
Comprises the following steps:
wherein
Representing the spectrum library to be cut in the k +1 iteration process
Column i, D(:,i)Indicates the ith column, H in the initialization spectral library D(:,i) (k+1)Column i, which represents the relaxation variable H during the (k + 1) th iteration, the soft function is the well-known soft threshold shrinkage function.
Step 8, outputting the cut spectral library and the end-member abundance map:
(1) output cut spectral library
(2) Matrix abundance coefficient
Restructuring into three-dimensional data
Then output
Wherein H represents the height of the mineral abundance map, W represents the width of the mineral abundance map, and N represents the spectral library after cutting
The number of end members in (1).
The real hyperspectral data set adopted in the simulation experiment of the embodiment is a custom data set shot by a Visible InfraRed Imaging Spectrometer (AVIRIS). The data used in the experiment was a portion of the Cuprite data, 250 x 191 pixels in size, containing 224 bands between 400 and 2500 nanometers, with a spectral resolution of 10 nanometers. Before the experiment, the wavelength bands 1-2, 105, 115, 150, 170, 223, 224 are removed due to the moisture absorption and the larger noise, and 188 usable wavelength bands remain. The simulation experiments are all completed by adopting MATLAB R2015b under a Windows 7 operating system.
The invention adopts the unmixing effect of a real hyperspectral data set inspection algorithm. In order to test the performance of the algorithm, the proposed hyperspectral data unmixing method (PSCSR) based on spectral library cutting and collaborative sparse regression is compared with the currently internationally popular unmixing algorithm. The comparison method comprises the following steps: the method of demixing built-in the software TECORDER, sparse demixing (SUnSAL) [ biological-Dias J M, Figueiredo M A T. alternation direction algorithm for constrained sparse regression [ WHISPERS ], 20102 kshoop. IEEE,2010:1-4 ], sparse mixing based on variable splitting and augmented Lagrange [ Iordhe M D, biological-Dias J M, plant A. collagen mapping for sparse regression [ biological D, biological-Dias J M, biological-As J, biological A. collagen mapping J ] and sparse regression [ sparse regression J ] model J.52. sparse regression, IEEE-1-4 ], sparse regression [ sparse regression J.S. III, III. J. (III) and sparse regression [ sparse J.S.: III, 2016,54(9):5171-5184.]
Fig. 2(a) to 2(o) are abundance diagrams of three minerals of alunite, hydroammonium feldspar and chalcedony under different unmixing algorithms of the Cuprite dataset. It can be seen that the abundance maps of the three minerals obtained by the unmixing algorithm provided by the invention are basically consistent with the reference map obtained by using the Tetracorder software, and the estimated chalcedony abundance map is closer to the reference map than the other three unmixing algorithms (SUnSAL, CLSUnSAL, DANSER), which proves that the unmixing algorithm can provide an accurate estimation for the real ground feature distribution. Fig. 3(a) -3 (c) are the comparison of the spectral curves of three minerals of alunite, diasphore feldspar and chalcedony in the initial spectral library and the cut spectral library. Fig. 3(a) shows that the difference between the alunite spectral curve and the cut alunite spectral curve occurs mostly at the absorption peak, that is, the spectral change of the same substance has certain sparsity.
Claims (3)
1. A hyperspectral data unmixing method based on spectral library cutting and collaborative sparse regression is characterized by comprising the following steps:
step 1, inputting a spectrum library and hyperspectral data to be unmixed; the specific process is as follows:
inputting a spectrum library containing hyperspectral end members
Wherein the integer K > 0 represents the number of bands of each spectral signal and the integer N>0 represents the number of end member spectra contained in the spectral library;
inputting hyperspectral data to be unmixed
Wherein the integer L>0 represents the number of bands of the hyperspectral data, an integer W>0 and an integer H > 0 represent the width and height of the hyperspectral data space dimension, respectively;
step 2, initializing a spectrum library and constructing a spectrum data matrix; the method comprises the following specific steps:
step 2-1, inputting a spectrum library end member
The number of wave bands is K > 0, the wave bands are
And inputting a hyperspectral signal to be unmixed
Number of wave bands L>0 Band ═ b1,b2,...,bL](ii) a According to the hyperspectral signal to be unmixed
Band of (1), reserve spectrum libraryEnd member
Wave band
The other corresponding wave bands are discarded; through the one-step operation, the end member signals in the spectral library
And the hyperspectral signal to be unmixed
The wave bands in the spectrum library are kept consistent to obtain the end members of the spectrum library after the wave bands are removed
Number of wave bands L>0 Band ═ b1,b2,...,bL]；
Step 2-2, inputting a spectrum library of
The number of wave bands is K > 0, the number of contained end members is N > 0, and for a spectrum library
The ith end-member signal of
Removing the wave bands by using the method in the step 2-1 to obtain the removed end member signals
Spectral library
Each end-member signal in (1) using the above methodThe method can obtain the initialized spectrum library
Step 2-3, initially inputting the hyperspectral data to be unmixed into
The hyperspectral data to be unmixed
Arranged pixel by pixel to form a spectral data matrix
Wherein the integer L>0 represents the number of wave bands, M is W multiplied by H represents the number of hyperspectral pixels,
representing a hyperspectral pixel, and taking a spectrum data matrix Y as one input of a model;
step 3, constructing a fitting error fidelity item between the hyperspectral data to be unmixed and the spectrum library matrix:
in the formula
Representing a hyperspectral data matrix to be unmixed,
representing the spectral library to be tailored,
is an abundance coefficient matrix, where L>0 represents the number of bands, M > 0 represents the hyperspectral dataThe number of spectrum signals in the matrix;
step 4, constructing an abundance matrix cooperative sparsity constraint term; the method specifically comprises the following steps:
in the formula
Representing a matrix of abundance coefficients, | · | | non-conducting phosphor2,1Is represented by2,1Norm, Ui,jRepresenting a hyperspectral data matrix to be unmixed
Spectrum library corresponding to jth pixel
The abundance coefficient of the ith end member is that j is more than or equal to 1 and less than or equal to M, and i is more than or equal to 1 and less than or equal to N;
step 5, constructing a cut sparse regularization item of the spectrum library; the method specifically comprises the following steps:
in the formula
Representing the spectral library after initialization,
representing the spectral library to be tailored,
to represent
To (1)p rows, the value corresponding to the q column,
represents the value corresponding to the p-th row and the q-th column in D, p is more than or equal to 1 and less than or equal to L, q is more than or equal to 1 and less than or equal to N, | | · | | survival1,1Is represented by1,1A norm;
step 6, establishing a spectral library cutting and collaborative sparse regression model; the method specifically comprises the following steps:
wherein the parameter lambda is more than 0, beta is more than 0,
representing a hyperspectral data matrix to be unmixed,
representing the spectral library after initialization,
representing the spectral library to be tailored,
representing a matrix of abundance coefficients, | · | | non-conducting phosphorFRepresenting the F norm of the matrix, | · | | non-woven phosphor2,1Is represented by2,1Norm, | · | luminance1,1Is represented by1,1A norm;
step 7, iteratively solving a spectral library cutting and collaborative sparse regression model;
and 8, outputting the cut spectral library and the end-member abundance map.
2. The hyperspectral data unmixing method based on spectral library cutting and collaborative sparse regression according to claim 1 is characterized in that the specific steps of iterating the spectral library cutting and collaborative sparse regression model in step 7 are as follows:
(1) inputting initialized spectral library
And the hyperspectral data matrix Y to be unmixed is [ Y ═ Y1,y2,...,yM]∈RL×M；
(2) Initializing an abundance coefficient matrix
And a spectral library to be tailored
(3) Introducing a relaxation variable
The spectral library cutting and collaborative sparse regression model is changed into the following model:
the parameters alpha is more than 0, lambda is more than 0, beta is more than 0, and the equality constraint U-V is introduced into the model1,U＝V2Then solving the augmented Lagrange function minimization model:
where the parameter μ > 0, variable
The abundance coefficient matrix can be obtained after the solution by using the exchange direction multiplier method
Comprises the following steps:
wherein U is(k+1)Represents the abundance matrix U, V in the k +1 iteration1 (k)Representing the variable V during the kth iteration1，D1 (k)Representing the Lagrange multiplier variable D during the kth iteration1，V2 (k)Representing the variable V during the kth iteration2，D2 (k)Representing the Lagrange multiplier variable D during the kth iteration2To find out the spectrum library to be cut
Comprises the following steps:
wherein,
representing the spectrum library to be cut in the k +1 iteration process
Column i, D(:,i)Indicates the ith column, H in the initialization spectral library D(:,i) (k+1)The ith column, representing the relaxation variable H during the (k + 1) th iteration, is the soft threshold shrinkage function.
3. The hyperspectral data unmixing method based on spectral library clipping and collaborative sparse regression according to claim 2, wherein the specific process of outputting the clipped spectral library and the end-member abundance map in step 8 is as follows:
(1) output cut spectral library
(2) Calculating a resulting abundance coefficient matrix using a tailored spectral library
Recombine it into three-dimensional data
Then output
Wherein H represents the height of the mineral abundance map, W represents the width of the mineral abundance map,
representation of spectral library after cutting
The number of end members in (1).
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810016715.4A CN108734672B (en) | 2018-01-08 | 2018-01-08 | Hyperspectral data unmixing method based on spectral library cutting and collaborative sparse regression |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810016715.4A CN108734672B (en) | 2018-01-08 | 2018-01-08 | Hyperspectral data unmixing method based on spectral library cutting and collaborative sparse regression |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108734672A CN108734672A (en) | 2018-11-02 |
| CN108734672B true CN108734672B (en) | 2022-02-18 |
Family
ID=63940371
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810016715.4A Active CN108734672B (en) | 2018-01-08 | 2018-01-08 | Hyperspectral data unmixing method based on spectral library cutting and collaborative sparse regression |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108734672B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109580495B (en) * | 2018-11-28 | 2021-08-24 | 重庆工商大学 | Demixing device and method based on hyperspectral image |
| CN111260576B (en) * | 2020-01-14 | 2022-07-05 | 哈尔滨工业大学 | Hyperspectral unmixing algorithm based on de-noising three-dimensional convolution self-coding network |
| CN113094645B (en) * | 2021-03-11 | 2024-07-30 | 南京理工大学 | Hyperspectral data unmixing method based on morphological component constraint optimization |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104463223A (en) * | 2014-12-22 | 2015-03-25 | 西安电子科技大学 | Hyperspectral image group sparse demixing method based on empty spectral information abundance restraint |
| CN105825227A (en) * | 2016-03-11 | 2016-08-03 | 南京航空航天大学 | Hyperspectral image sparseness demixing method based on MFOCUSS and low-rank expression |
| US20160371563A1 (en) * | 2015-06-22 | 2016-12-22 | The Johns Hopkins University | System and method for structured low-rank matrix factorization: optimality, algorithm, and applications to image processing |
-
2018
- 2018-01-08 CN CN201810016715.4A patent/CN108734672B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104463223A (en) * | 2014-12-22 | 2015-03-25 | 西安电子科技大学 | Hyperspectral image group sparse demixing method based on empty spectral information abundance restraint |
| US20160371563A1 (en) * | 2015-06-22 | 2016-12-22 | The Johns Hopkins University | System and method for structured low-rank matrix factorization: optimality, algorithm, and applications to image processing |
| CN105825227A (en) * | 2016-03-11 | 2016-08-03 | 南京航空航天大学 | Hyperspectral image sparseness demixing method based on MFOCUSS and low-rank expression |
Non-Patent Citations (2)
| Title |
|---|
| Joint Local Abundance Sparse Unmixing for Hyperspectral Images;Mia Rizkinia 等;《Remote Sensing》;20171127;第1-22页 * |
| Semiblind Hyperspectral Unmixing in the Presence of Spectral Library Mismatches;Xiao Fu 等;《IEEE Transactions on Geoscience and Remote Sensing》;20160510;第5171-5184页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108734672A (en) | 2018-11-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111260576B (en) | Hyperspectral unmixing algorithm based on de-noising three-dimensional convolution self-coding network | |
| CN104978573B (en) | A kind of non-negative matrix factorization method applied to Hyperspectral imagery processing | |
| CN104952050B (en) | High spectrum image adaptive de mixing method based on region segmentation | |
| CN107292317B (en) | Polarization SAR classification method based on shallow feature and T matrix deep learning | |
| CN108734672B (en) | Hyperspectral data unmixing method based on spectral library cutting and collaborative sparse regression | |
| CN107292258B (en) | High-spectral image low-rank representation clustering method based on bilateral weighted modulation and filtering | |
| CN104182978B (en) | A kind of high spectrum image object detection method based on empty spectrum nuclear sparse expression | |
| CN106447630B (en) | The high spectrum image sharpening method decomposed based on probability matrix | |
| CN104463223B (en) | Hyperspectral image group sparse unmixing method based on space spectrum information abundance constraint | |
| CN102540271B (en) | Semi-supervised hyperspectral sub-pixel target detection method based on enhanced constraint sparse regression method | |
| CN105825227B (en) | A kind of high spectrum image sparse solution mixing method based on MFOCUSS and low-rank representation | |
| CN104933425B (en) | A kind of hyperspectral data processing method | |
| CN112712034B (en) | Unmixing method and system for hyperspectral image | |
| CN111680579B (en) | Remote sensing image classification method for self-adaptive weight multi-view measurement learning | |
| CN104268561B (en) | High spectrum image solution mixing method based on structure priori low-rank representation | |
| CN103871087A (en) | Hyperspectral unmixing compressive sensing method based on three-dimensional total variation sparse prior | |
| CN103903261A (en) | Spectrum image processing method based on partition compressed sensing | |
| CN108470192B (en) | Hyperspectral classification method and device | |
| CN105184314A (en) | wrapper-type hyperspectral waveband selection method based on pixel clustering | |
| CN117058522A (en) | Remote sensing earth surface ecological change detection method integrating spectral neural network and leaf area index | |
| CN115144357A (en) | Mineral identification method based on airborne mid-infrared hyperspectral remote sensing data of mining area | |
| CN110674848A (en) | High-dimensional data joint clustering method combining sparse representation and bipartite graph segmentation | |
| CN108280486A (en) | A kind of high spectrum image solution mixing method based on end member cluster | |
| CN105427351B (en) | Compression of hyperspectral images cognitive method based on manifold structure sparse prior | |
| CN105354584B (en) | High-spectral data wave band based on wave band dissimilarity characterizes selection method |
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 |