CN114120020A - Hyperspectral image inter-spectrum sequencing method based on key channel protection and spectral clustering - Google Patents
Hyperspectral image inter-spectrum sequencing method based on key channel protection and spectral clustering Download PDFInfo
- Publication number
- CN114120020A CN114120020A CN202111449378.6A CN202111449378A CN114120020A CN 114120020 A CN114120020 A CN 114120020A CN 202111449378 A CN202111449378 A CN 202111449378A CN 114120020 A CN114120020 A CN 114120020A
- Authority
- CN
- China
- Prior art keywords
- channel
- channels
- matrix
- key
- spectrum
- 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.)
- Granted
Links
- 238000001228 spectrum Methods 0.000 title claims abstract description 56
- 230000003595 spectral effect Effects 0.000 title claims abstract description 43
- 238000012163 sequencing technique Methods 0.000 title claims description 26
- 238000000034 method Methods 0.000 claims abstract description 72
- 238000007906 compression Methods 0.000 claims abstract description 35
- 230000006835 compression Effects 0.000 claims abstract description 34
- 239000011159 matrix material Substances 0.000 claims description 86
- 238000013507 mapping Methods 0.000 claims description 6
- 230000011218 segmentation Effects 0.000 claims description 3
- 239000013598 vector Substances 0.000 claims description 3
- 238000012545 processing Methods 0.000 abstract description 7
- 230000005540 biological transmission Effects 0.000 description 14
- 238000001514 detection method Methods 0.000 description 11
- 238000004364 calculation method Methods 0.000 description 7
- 238000013144 data compression Methods 0.000 description 5
- 230000000694 effects Effects 0.000 description 5
- 230000008520 organization Effects 0.000 description 5
- 238000012216 screening Methods 0.000 description 5
- 238000010276 construction Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 230000003111 delayed effect Effects 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 230000000750 progressive effect Effects 0.000 description 2
- 230000005855 radiation Effects 0.000 description 2
- YGULWPYYGQCFMP-CEAXSRTFSA-N Metoprolol tartrate Chemical class OC(=O)[C@H](O)[C@@H](O)C(O)=O.COCCC1=CC=C(OCC(O)CNC(C)C)C=C1.COCCC1=CC=C(OCC(O)CNC(C)C)C=C1 YGULWPYYGQCFMP-CEAXSRTFSA-N 0.000 description 1
- 238000010521 absorption reaction Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 230000006837 decompression Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000000704 physical effect Effects 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
- 230000002441 reversible effect Effects 0.000 description 1
- 238000010187 selection method Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/231—Hierarchical techniques, i.e. dividing or merging pattern sets so as to obtain a dendrogram
-
- 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
-
- 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
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
A hyperspectral image inter-spectrum ordering method based on key channel protection and spectral clustering relates to the technical field of image processing, and aims to solve the problem that the prior hyperspectral image compression processing flow can not simultaneously meet the requirements of key channels which are transmitted preferentially and play an important role in application and the requirements of predicting and compressing the residual channels by using the information of the key channels. The method and the device can meet the requirement of preferentially transmitting the key channel which plays an important role in application and the requirement of predicting and compressing the residual channels by using the information of the key channel at the same time.
Description
Technical Field
The invention relates to the technical field of image processing, in particular to a hyperspectral image inter-spectrum sequencing method based on key channel protection and spectral clustering.
Background
Infrared atmosphere detection interferometer (IASI) is an important instrument for Numerical Weather Prediction (NWP). The method uses 8461 channels outside the atmosphere to measure atmospheric radiation, the spatial resolution of each channel is 60 multiplied by 1530 multiplied by 16bits, the data volume of an IASI hyperspectral image is about 1.45Gbytes, and a large amount of data generated by a hyperspectral infrared detector such as the IASI presents a plurality of challenges, particularly in the aspects of data storage, calculation cost, information redundancy, information content and the like. In addition, the physical inversion of the hyperspectral image involves solving a radiation transmission integral equation, the equation usually has a nonlinear problem, and tiny data disturbance can bring huge influence, so that the volume of the hyperspectral image needs to be reduced by lossless compression, but the compression ratio of the lossless compression is limited. At present, the requirement of preferentially transmitting a key channel which plays an important role in application and the requirement of predicting and compressing the residual channels by using the information of the key channel cannot be simultaneously met in a hyperspectral image compression processing flow.
Disclosure of Invention
The purpose of the invention is: aiming at the problem that the prior hyper-spectral image compression processing flow can not simultaneously meet the requirements of preferentially transmitting key channels which have important effects on application and the requirements of predicting and compressing the residual channels by using the information of the key channels, the method for sequencing the hyper-spectral images among spectrums based on key channel protection and spectrum clustering is provided.
The technical scheme adopted by the invention to solve the technical problems is as follows:
the hyperspectral image inter-spectrum sequencing method based on key channel protection and spectral clustering comprises the following steps:
the method comprises the following steps: acquiring a linear correlation matrix R between spectrums;
step two: weighting and normalizing the linear correlation matrix R among the spectrums according to the physical segmentation characteristics to obtain a similarity matrix W;
step three: selecting key channels of interest;
step four: obtaining channel groups by adopting a hierarchical clustering method according to the similarity matrix W, then setting a threshold value, setting channels in the groups with the number of the channels smaller than the threshold value as specific channels, and then subtracting the key channels and the specific channels from all the channels to obtain common channels;
step five: respectively selecting representative channels of the key channel, the specific channel and the common channel according to the similarity matrix W to obtain a representative channel of the key channel, a representative channel of the specific channel and a representative channel of the common channel;
step six: outputting spectrum sequencing according to the sequence of the representative channel of the key channel, the representative channel of the common channel, the key channel, the common channel, the representative channel of the specific channel and the specific channel;
step seven: and (5) finishing hyper-spectrum compression by utilizing the spectrum sequencing.
Further, the linear correlation matrix R between the spectra obtained in the first step is obtained by Pearson linear correlation coefficients.
Further, the distance measure between classes in the hierarchical clustering method adopts the nearest distance.
Furthermore, the key channel in the third step is a temperature and humidity sensitive channel.
Further, the temperature and humidity sensitive channel is obtained by iteratively selecting a spectrum channel with the information content ratio reaching 99% through the information content.
Further, the inter-spectrum linear correlation matrix R is represented as:
wherein f isi(x, y) and fj(x, y) is the pixel gray scale value, μ, of the ith and jth channels at spatial location (x, y)iAnd mujIs the mean value of the gray values of the ith and jth channel images, RijRepresents the linear correlation of the ith channel and the jth channel, and M and N are single-channel images fiNumber of rows and columns, muiAnd mujRespectively expressed as:
further, the similarity matrix W is represented as:
where max () represents the maximum value of all matrix elements, WijRepresenting the correlation of the jth channel of the ith channel weighted by the prior physical segment characteristic information.
Further, the specific steps of respectively selecting the representative channels of the key channel, the specific channel and the common channel according to the similarity matrix W in the step five are as follows:
a1: constructing a degree matrix D, the degree matrix D being represented as:
D=diag(d1,d2...dn)
wherein diag () denotes that the diagonal matrix is constructed by the elements in parentheses, i is 1. ltoreq. n,represents to WsIs summed column by column, j denotes a row, WsThe method comprises the steps that a submatrix is formed by similarity matrixes W of input channel sets to be subjected to spectral clustering grouping at corresponding positions, wherein the channel sets comprise key channels, common channels and specific channels;
a2: according to degree matrices D and WsConstructing an unnormalized Laplace matrix, wherein the unnormalized Laplace matrix is expressed as:
L=D-Ws;
a3: constructing a normalized Laplace matrix according to the non-normalized Laplace matrix and the degree matrix D, wherein the normalized Laplace matrix is expressed as follows:
a4: calculating LnorMinimum k negationsForming a feature matrix F according to the feature vector F corresponding to the zero eigenvalue and standardization of columns, wherein each row of the F is used as a k-dimensional sample;
a5: clustering the feature matrix F by using a K central point method to obtain a cluster center C (C)1,c2,…,ck);
A6: and mapping the cluster center back to the channel serial number to obtain a representative channel.
Further, the specific steps of respectively selecting the representative channels of the key channel, the specific channel and the common channel according to the similarity matrix W in the step five are as follows:
b1: constructing a degree matrix D, the degree matrix D being represented as:
D=diag(d1,d2...dn)
wherein diag () denotes that the diagonal matrix is constructed by the elements in parentheses, i is 1. ltoreq. n,represents to WsIs summed column by column, j denotes a row, WsThe method comprises the steps that a submatrix is formed by similarity matrixes W of input channel sets to be subjected to spectral clustering grouping at corresponding positions, wherein the channel sets comprise key channels, common channels and specific channels;
b2: according to degree matrices D and WsConstructing a normalized Laplace matrix, wherein the normalized Laplace matrix is expressed as:
wherein, WsThe method comprises the steps that a submatrix is formed by similarity matrixes W of input channel sets to be subjected to spectral clustering grouping at corresponding positions, wherein the channel sets comprise key channels, common channels and specific channels;
b3: calculating LnorMaximum k eigenvalue corresponding featuresThe vector F is normalized according to columns to form a characteristic matrix F, and each row of the characteristic matrix F is used as a k-dimensional sample;
b4: clustering the characteristic matrix F by using a K central point method to obtain a cluster center;
b5: and mapping the cluster center back to the channel serial number to obtain a representative channel.
The invention has the beneficial effects that:
1) the representative channels of the key channels and the representative channels of the common channels are arranged at the top, the representative channels are obtained from all groups of clustering grouping centers, the average similarity between the clustering centers and the clustering groups of the clustering centers is the maximum, and other channels in the subsequent clustering groups can be well predicted.
2) Although the key channel is arranged at the third position for transmission, the representative channel is obtained from the cluster grouping centers of each group, and the number of the representative channels is very small, so that the priority transmission of the key channel is still ensured, and the lossless compression ratio is further improved.
3) Because the transmission of the spectrum channels is gradual, for the subsequent prediction compression, the prediction capability of the predictor becomes stronger along with the increase of the number of the transmission channels, and the specific channels are placed at the tail end of the last sequencing bit, so that the predictor with stronger capability can be ensured to predict the specific channels which are difficult to predict, and the overall prediction level is improved.
4) The method and the device can meet the requirement of preferentially transmitting the key channel which plays an important role in application and the requirement of predicting and compressing the residual channels by using the information of the key channel at the same time.
Drawings
FIG. 1 is a schematic representation of the infrared spectral absorption bands of the major gas components in the IASI detection range;
FIG. 2 is a schematic diagram showing the distribution of 371 temperature inversion sensitive channels selected according to the information quantity step-by-step iteration method in the detection band;
FIG. 3 is a schematic distribution diagram of 246 humidity inversion sensitive channels selected according to the information quantity step-by-step iteration method in the detection band;
FIG. 4 is a schematic distribution diagram of 615 temperature and humidity inversion sensitive channels selected according to an information quantity step-by-step iteration method as key channels in a detection waveband;
FIG. 5 shows a channel ordering composition diagram;
fig. 6 is a general flow chart of the present application.
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: referring to fig. 1, the method for sorting among spectra of a hyperspectral image based on key channel protection and spectral clustering according to the embodiment is specifically described, and the method comprises the following steps:
the method comprises the following steps: acquiring a linear correlation matrix R between spectrums;
step two: weighting and normalizing the linear correlation matrix R among the spectrums according to the physical segmentation characteristics to obtain a similarity matrix W;
step three: selecting key channels of interest;
step four: obtaining channel groups by adopting a hierarchical clustering method according to the similarity matrix W, then setting a threshold value, setting channels in the groups with the number of the channels smaller than the threshold value as specific channels, and then subtracting the key channels and the specific channels from all the channels to obtain common channels;
step five: respectively selecting representative channels of the key channel, the specific channel and the common channel according to the similarity matrix W to obtain a representative channel of the key channel, a representative channel of the specific channel and a representative channel of the common channel;
step six: outputting spectrum sequencing according to the sequence of the representative channel of the key channel, the representative channel of the common channel, the key channel, the common channel, the representative channel of the specific channel and the specific channel;
step seven: and (5) finishing hyper-spectrum compression by utilizing the spectrum sequencing.
At present, channel selection methods are mature, one of the most widely used methods is the theory from Rodgers, the theory describes an iterative method for determining an optimal channel set based on the information content, single-channel information content is calculated through an error covariance matrix before observation, a weight function matrix and an observation error matrix, and then the information content of a selected channel is maximized in a step-by-step iteration mode to perform channel selection, so that the method selects a temperature and humidity sensitive channel as a key channel by taking the Rodgers theory as an example. It is noted that the selection of the key channel is flexible, that is, the corresponding channel combination can be selected as the key channel in different application scenarios.
The linear prediction method is a common decorrelation method, the hyperspectral image prediction compression method based on orthogonal matching pursuit is adopted, a webp and jpeg2000 encoder is used for removing spatial dimension correlation, and an interval encoder is used for entropy encoding. The method has the advantages that the spectral channels are mainly sorted, so that the specific decorrelation and compression method can be replaced by other hyperspectral or hyperspectral lossless data compression methods with inter-spectrum linear prediction.
The method combines a meteorological hyperspectral image channel selection theory with a lossless data compression theory of an image processing subject, and provides a hyperspectral image inter-spectrum sequencing algorithm which takes a spectrum channel with large contribution to inversion application and prediction compression as a key channel for preferential transmission and a spectrum channel with small contribution to prediction compression for delayed transmission. The algorithm can realize progressive transmission of data in the process of transmission and decompression of the hyperspectral images, ensures that key channels which are very important for application are transmitted preferentially, provides good inversion quality with few key channels, ensures quick and flexible application of the hyperspectral images, simultaneously, the key channels are also used for predicting subsequent channels, and all lossless hyperspectral images can be obtained after transmission and decoding with the subsequent channels are finished.
Aiming at the problem that the requirements of preferentially transmitting key channels having important functions on application cannot be met simultaneously in the conventional hyperspectral image compression processing flow and the requirements of predicting and compressing the residual channels by using the information of the key channels, the hyperspectral image spectrum sequencing algorithm based on key channel protection and spectral clustering is provided, and the algorithm only generates the arrangement sequence of the hyperspectral image spectrums, so the hyperspectral image spectrum sequencing algorithm is reversible. By the sequencing algorithm, the key channel is preferentially transmitted to meet the application requirement, and the compression ratio of the hyperspectral image is slightly improved. The method comprises the following specific steps:
the method comprises the following steps: correlation calculation
Step two: correlation coefficient matrix weighting
Step three: determining critical channels
Step four: screening for specific channels
Step five: selecting representative channel
Step six: organization gets ranked
Step seven: compression using inter-spectral prediction
The first step, the second step, the third step and the fourth step are main bodies of the method, the first step, the second step and the third step are used for generating channel sequencing, the seventh step is a subsequent prediction compression process, a hyperspectral image prediction compression method based on orthogonal matching pursuit is used, and the hyperspectral or hyperspectral lossless data compression method with the linear prediction among spectrums can be replaced by other hyperspectral or hyperspectral lossless data compression methods. The following description is directed to core steps one through six.
The method comprises the following steps: inter-spectral correlation calculation
The inter-spectrum ordering of the hyper-spectrum image needs to meet application requirements on one hand, and on the other hand, the inter-spectrum prediction effect is better through the ordering, namely, the spectrum channel ordered in front can well represent the spectrum channel behind the spectrum channel. There is therefore a need for a way to measure the correlation between spectral channels. Considering that linear prediction is used in the prediction stage, correlation is calculated using Pearson linear correlation number, and a larger correlation coefficient indicates a higher linear correlation.
Step two: correlation coefficient matrix weighting
The satellite infrared atmosphere vertical detection technology is developed from a filter type to a grating type and an interference type, the number of channels is increased from dozens to thousands, the higher spectral resolution enables a hyper-spectral image to reflect the physical characteristics of atmospheric components better, even if the channel numbers of the spectral channels affected by the same gas component are greatly different, the spectral channels have strong correlation, and therefore judgment needs to be carried out by combining the physical characteristics of the atmosphere and the data characteristics of the hyper-spectral image. Taking the IASI data as an example, table 1 introduces the detection applications of different spectral channels.
TABLE 1 detection application of different wave bands of hyper-spectral detector
The channel correlation in the same segment is weighted, and since the application only concerns the correlation strength and does not consider positive correlation or negative correlation, the absolute value of the correlation coefficient is required to obtain the similarity matrix.
Step three: determining critical channels
The key channel can be arbitrarily designated as required, here, a spectrum channel with an information content ratio of 99% is iteratively selected as the key channel through an information content by taking a temperature and humidity sensitive channel as an example, and fig. 1 to fig. 3 show the selection result of the key channel. Fig. 2 is the distribution of 371 temperature inversion sensitive channels selected according to the information quantity step iteration method under the detection wave band, and fig. 3 is the distribution of 246 humidity inversion sensitive channels selected according to the information quantity step iteration method under the detection wave band.
Step four: screening for specific channels
For most channels, they are relatively strong to adjacent channels, but for some specific spectral channels, they are relatively poor to most spectral channels, which we call specific channels, these specific channels do not contribute much to the prediction of the subsequent channels, if they are sorted earlier, they will affect the prediction of the subsequent channels, so it is necessary to put their delayed transmission at the end of the sorting.
And for the selection of the specific channel, a hierarchical clustering method is adopted, the distance measurement between classes adopts the nearest distance, and the channels in the groups smaller than the threshold value are set as the specific channels. From the point of view of graph theory, this hierarchical clustering operation into k groups is equivalent to generating a minimum generated forest with k trees, while the idiosyncratic channels are the set of nodes of those trees that are too small.
Step five: selecting representative channel
Through the third and fourth steps, the key channels and the specific channels are obtained, and the number of the rest channels is called as the common channels, in order to better predict and express the common channels, some channels are selected as representatives to be better predicted, a means of spectral clustering is applied to obtain the representative channels through k central point clustering, although hierarchical clustering can still be applied here to generate representative channels, hierarchical clustering at step four results in minimal forest generation, whereas tree or forest requirements are loop-free, but for predictive operation, the presence of loops is advantageous, since more weighted edges mean more similar channels are available to participate in the prediction, therefore, a graph formed by taking the channel as a node is cut into a plurality of connected branches by adopting a graph cutting method of spectral clustering, and the clustering center of a point set in each connected branch is taken as a representative channel.
Usually, the spectral clustering adopts a Gaussian kernel function to construct a similarity matrix, but in consideration of the physical properties of the hyperspectral images, a graph is constructed by using the similarity matrix W in the second step. Taking each channel image as a vertex to form a weighted undirected graph G ═ (V, E), V ═ V1,v2,…,vnV is a spectral channel, E ═ E1,e2,…,emOn the edge WijCalculated from (3), the degree of the vertex is recorded as diThe sum of degrees of the vertices of one connected branch is vol ().
vol(A)=∑i∈Adi (1)
Spectral clustering is to minimize the following objective function:
clustering centers can be selected from the key channel and the specific channel through spectral clustering, and the clustering centers are used as representative channels.
Step six: organization gets ranked
And outputting spectrum sequencing according to the sequence of the representative channel of the key channel, the representative channel of the common channel, the key channel, the common channel, the representative channel of the specific channel and the specific channel, and sequencing the channels in each section from small to large according to the sequence numbers.
The method for guaranteeing the priority transmission of key channels which are very important for application can be obviously seen from the sixth step of the execution step, in order to check the improvement of the lossless compression ratio of the key channel protection and spectral clustering-based hyperspectral image inter-spectrum sequencing algorithm, a hyperspectral image prediction compression method based on orthogonal matching pursuit is selected as a compression mode, the lossless compression ratio before and after sequencing is explained, and in addition, the method is compared with the multi-stage clustering RKLT + M-CALIC method with the highest data compression ratio at present, and the compression ratio reaches an advanced level.
The experimental data come from L1C grade data (60 lines x 1530 column x 8461 channels x 16bits) of IASI detector carried on METOP series satellites of European meteorological satellite development organization, the experiments losslessly compressed and encoded data of 15 different orbits at different times, and the data come from data center of European meteorological satellite application organization (http:// catalog.ceda.ac.uk /)
TABLE 1 Hyperspectral detection data specific information and compression ratio of 15 IASI complete orbits used in the experiment
The compression ratio is calculated as (original data size)/(compressed data size), and the larger the compression ratio is, the better the compression effect is. When the upper limit of the number of the parameter reference channels S is 9 and the number of the spatial clustering groups k is 8, the average compression ratio is 2.5418, and is improved by 0.0032 compared with the average compression ratio 2.5386 without sequencing, the lossless compression effect of the method is equivalent to the 2.54 compression ratio of the current multistage clustering RKLT + M-CALIC method with the best compression effect, and progressive data transmission is supported, and screening and preferential transmission of key channels are supported.
Example (b):
the overall process flow is described in conjunction with fig. 6 as follows:
the method comprises the following steps: inter-spectral correlation calculation
Considering that linear prediction is used in the prediction stage, correlation is calculated using Pearson linear correlation coefficients, and a larger correlation coefficient indicates a higher linear correlation.
Wherein f isi(x, y) and fj(x, y) is the pixel gray scale value, μ, of the ith and jth channels at spatial location (x, y)iAnd mujThe calculation formula is the gray value average value of the ith channel image and the jth channel image as follows:
the linear correlation between any two channels of the hyperspectral image can be calculated through the formula (4), and a correlation coefficient matrix R is formed.
Step two: correlation coefficient matrix weighting
The hyperspectral image can be divided into 11 segments in the spectral dimension according to the physical characteristics of the hyperspectral, which is as follows:
TABLE 3 grouping based on IASI physical characteristics
For the first 10 segments in table 2, weighting the channel correlation in the same segment, and taking the absolute value of the correlation coefficient to obtain the similarity matrix W since we only care about the correlation strength and do not consider whether it is positive or negative correlation
According to experience, r is generally 0.1, where | | | represents taking an absolute value, and for convenience of subsequent calculation, W needs to be normalized
Where max () represents the maximum of all matrix elements.
Step three: determining critical channels
The key channel can be arbitrarily specified according to requirements, and a spectrum channel with the information content ratio of 99% is selected as the key channel by taking a step-by-step iteration method which is provided by Rodgers and utilizes prior information to calculate the information entropy as an example.
Step four: screening for specific channels
And taking the similarity matrix obtained in the step two as similarity measurement among spectra, taking the nearest distance as an inter-class distance standard, dividing 8461 spectral channels of the IASI into 70 groups, screening groups with less than 10 channels in the groups, and marking the channels as specific channels.
Step five: selecting representative channel
There are two ways to realize spectral clustering, the two ways are equivalent, but for computer realization, because the second way is to find the maximum eigenvalue, the calculation speed will be faster than the first way.
The first method is as follows:
1) the construction degree matrix D, diag () represents the construction of a diagonal matrix from the parenthesized elements.
D=diag(d1,d2...dn) (8)
2) Constructing an unnormalized Laplace matrix representation as:
L=D-W (9)
3) the constructed normalized Laplace matrix is:
4) calculating LnorAnd (3) normalizing the eigenvector F corresponding to the minimum k non-zero eigenvalues by columns to form an eigenvector matrix F, wherein each row of the F is used as a k-dimensional sample.
5) Clustering using the K center point method to obtain cluster centers C (C)1,c2,…,ck)。
6) And mapping the cluster center back to the channel serial number to obtain a representative channel.
The second method comprises the following steps:
1) the construction degree matrix D, diag () represents the construction of a diagonal matrix from the parenthesized elements.
D=diag(d1,d2...dn) (8)
2) The constructed normalized Laplace matrix is:
4) calculating LnorThe maximum k eigenvalues correspond to the eigenvector F and are normalized by columns to form a feature matrix F, and each row of F is used as a k-dimensional sample.
5) And clustering the F by using a K center point method to obtain a cluster center.
6) And mapping the cluster center back to the channel serial number to obtain a representative channel.
4, 40 and 2 representative channels can be respectively selected from the key channel, the common channel and the specific channel through spectral clustering.
Step six: organization gets ranked
And outputting spectrum sequencing according to the sequence of the representative channel of the key channel, the representative channel of the common channel, the key channel, the common channel, the representative channel of the specific channel and the specific channel, and sequencing the channels in each section from small to large according to the sequence numbers. Fig. 5 shows the composition of the ordering.
Step seven: and obtaining a compressed code stream by a hyperspectral image prediction compression method based on orthogonal matching pursuit.
And the sequenced hyperspectral images can be decompressed in a lossless manner at a decoding end according to the compressed code stream, and the hyperspectral images can be rearranged according to the original sequence by transmitting the sequencing index to obtain lossless recovery.
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 (9)
1. The hyperspectral image inter-spectrum sequencing method based on key channel protection and spectral clustering is characterized by comprising the following steps of:
the method comprises the following steps: acquiring a linear correlation matrix R between spectrums;
step two: weighting and normalizing the linear correlation matrix R among the spectrums according to the physical segmentation characteristics to obtain a similarity matrix W;
step three: selecting key channels of interest;
step four: obtaining channel groups by adopting a hierarchical clustering method according to the similarity matrix W, then setting a threshold value, setting channels in the groups with the number of the channels smaller than the threshold value as specific channels, and then subtracting the key channels and the specific channels from all the channels to obtain common channels;
step five: respectively selecting representative channels of the key channel, the specific channel and the common channel according to the similarity matrix W to obtain a representative channel of the key channel, a representative channel of the specific channel and a representative channel of the common channel;
step six: outputting spectrum sequencing according to the representative channel of the key channel, the representative channel of the common channel, the key channel, the common channel, the representative channel of the specific channel and the sequence of the specific channel;
step seven: and (5) finishing hyper-spectrum compression by utilizing the spectrum sequencing.
2. The hyperspectral image inter-spectrum ordering method based on key channel protection and spectral clustering according to claim 1 is characterized in that the inter-spectrum linear correlation matrix R obtained in the first step is obtained by Pearson linear correlation coefficients.
3. The hyperspectral image inter-spectral ordering method based on key channel protection and spectral clustering as claimed in claim 2, wherein the distance measure between classes in the hierarchical clustering method adopts the nearest distance.
4. The hyperspectral image inter-spectrum ordering method based on key channel protection and spectral clustering according to claim 3, characterized in that the key channel in the third step is a temperature and humidity sensitive channel.
5. The hyperspectral image inter-spectrum ordering method based on key channel protection and spectral clustering according to claim 4 is characterized in that the temperature and humidity sensitive channel is obtained by iteratively selecting a spectral channel with the information content ratio reaching 99% through the information content.
6. The method for ordering spectrally based on key channel protection and spectral clustering of hyperspectral images as claimed in claim 5, wherein the linear correlation matrix R between spectra is represented as:
wherein f isi(x, y) and fj(x, y) is the pixel gray scale value, μ, of the ith and jth channels at spatial location (x, y)iAnd mujIs the mean value of the gray values of the ith and jth channel images, RijRepresents the linear correlation of the ith channel and the jth channel, and M and N are single-channel images fiNumber of rows and columns, muiAnd mujRespectively expressed as:
7. the hyperspectral image inter-spectrum ordering method based on key channel protection and spectral clustering according to claim 6, wherein the similarity matrix W is represented as:
where max () represents the maximum value of all matrix elements, WijRepresenting the correlation of the jth channel of the ith channel weighted by the prior physical segment characteristic information.
8. The hyperspectral image inter-spectrum ordering method based on key channel protection and spectral clustering according to claim 7 is characterized in that the concrete steps of respectively selecting the representative channels of the key channel, the specific channel and the common channel according to the similarity matrix W in the fifth step are as follows:
a1: constructing a degree matrix D, the degree matrix D being represented as:
D=diag(d1,d2...dn)
wherein diag () denotes that the diagonal matrix is constructed by the elements in parentheses, i is 1. ltoreq. n,represents to WsIs summed column by column, j denotes a row, WsThe method comprises the steps that a submatrix is formed by similarity matrixes W of input channel sets to be subjected to spectral clustering grouping at corresponding positions, wherein the channel sets comprise key channels, common channels and specific channels;
a2: according to degree matrices D and WsConstructing an unnormalized Laplace matrix, wherein the unnormalized Laplace matrix is expressed as:
L=D-Ws;
a3: constructing a normalized Laplace matrix according to the non-normalized Laplace matrix and the degree matrix D, wherein the normalized Laplace matrix is expressed as follows:
a4: calculating LnorForming a feature matrix F by normalizing feature vectors F corresponding to the minimum k non-zero eigenvalues according to columns, wherein each row of the F is used as a k-dimensional sample;
a5: clustering the feature matrix F by using a K central point method to obtain a cluster center C (C)1,c2,…,ck);
A6: and mapping the cluster center back to the channel serial number to obtain a representative channel.
9. The hyperspectral image inter-spectrum ordering method based on key channel protection and spectral clustering according to claim 1 is characterized in that the concrete steps of respectively selecting the representative channels of the key channel, the specific channel and the common channel according to the similarity matrix W in the fifth step are as follows:
b1: constructing a degree matrix D, the degree matrix D being represented as:
D=diag(d1,d2...dn)
wherein diag () denotes that the diagonal matrix is constructed by the elements in parentheses, i is 1. ltoreq. n,represents to WsIs summed column by column, j denotes a row, WsThe method comprises the steps that a submatrix is formed by similarity matrixes W of input channel sets to be subjected to spectral clustering grouping at corresponding positions, wherein the channel sets comprise key channels, common channels and specific channels;
b2: according to degree matrices D and WsConstructing a normalized Laplace matrix, wherein the normalized Laplace matrix is expressed as:
wherein, WsThe method comprises the steps that a submatrix is formed by similarity matrixes W of input channel sets to be subjected to spectral clustering grouping at corresponding positions, wherein the channel sets comprise key channels, common channels and specific channels;
b3: calculating LnorMaximum k eigenvalues correspond to the eigenvector F and are normalized according to columns to form an eigenvector matrix F, and each row of the F is used as a k-dimensional sample;
b4: clustering the characteristic matrix F by using a K central point method to obtain a cluster center;
b5: and mapping the cluster center back to the channel serial number to obtain a representative channel.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111449378.6A CN114120020B (en) | 2021-11-30 | 2021-11-30 | Method for ordering hyperspectral images among spectrums based on key channel protection and spectrum clustering |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111449378.6A CN114120020B (en) | 2021-11-30 | 2021-11-30 | Method for ordering hyperspectral images among spectrums based on key channel protection and spectrum clustering |
Publications (2)
Publication Number | Publication Date |
---|---|
CN114120020A true CN114120020A (en) | 2022-03-01 |
CN114120020B CN114120020B (en) | 2024-04-26 |
Family
ID=80369245
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111449378.6A Active CN114120020B (en) | 2021-11-30 | 2021-11-30 | Method for ordering hyperspectral images among spectrums based on key channel protection and spectrum clustering |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114120020B (en) |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105374054A (en) * | 2015-11-17 | 2016-03-02 | 重庆邮电大学 | Hyperspectral image compression method based on spatial spectrum characteristics |
CN109089118A (en) * | 2018-09-26 | 2018-12-25 | 哈尔滨工业大学 | Ultraphotic based on key message protection composes atmosphere infrared remote sensing method for compressing image |
WO2020155755A1 (en) * | 2019-01-28 | 2020-08-06 | 平安科技(深圳)有限公司 | Spectral clustering-based optimization method for anomaly point ratio, device, and computer apparatus |
AU2020103887A4 (en) * | 2020-12-04 | 2021-02-11 | kale, Karbhari Vishwanath DR | A method for automated endmember identification, selection and extraction from hyperspectral imagery |
-
2021
- 2021-11-30 CN CN202111449378.6A patent/CN114120020B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105374054A (en) * | 2015-11-17 | 2016-03-02 | 重庆邮电大学 | Hyperspectral image compression method based on spatial spectrum characteristics |
CN109089118A (en) * | 2018-09-26 | 2018-12-25 | 哈尔滨工业大学 | Ultraphotic based on key message protection composes atmosphere infrared remote sensing method for compressing image |
WO2020155755A1 (en) * | 2019-01-28 | 2020-08-06 | 平安科技(深圳)有限公司 | Spectral clustering-based optimization method for anomaly point ratio, device, and computer apparatus |
AU2020103887A4 (en) * | 2020-12-04 | 2021-02-11 | kale, Karbhari Vishwanath DR | A method for automated endmember identification, selection and extraction from hyperspectral imagery |
Non-Patent Citations (2)
Title |
---|
王立国;张晔;陈浩;: "基于鲁棒支持向量机的光谱解译", 吉林大学学报(工学版), no. 01, 10 January 2007 (2007-01-10) * |
陈善学;胡灿;屈龙瑶;: "基于自适应波段聚类PCA的高光谱图像压缩", 科学技术与工程, no. 12, 28 April 2015 (2015-04-28) * |
Also Published As
Publication number | Publication date |
---|---|
CN114120020B (en) | 2024-04-26 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Meller et al. | Same, same but different: Recovering neural network quantization error through weight factorization | |
Fu et al. | Adaptive spectral–spatial compression of hyperspectral image with sparse representation | |
CN108446599B (en) | Hyperspectral image band rapid selection method of p-value statistical modeling independence | |
CN109087367B (en) | High-spectrum image rapid compressed sensing reconstruction method based on particle swarm optimization | |
CN112067129B (en) | Hyperspectral processing method and waveband selection method | |
Gillis et al. | A fast gradient method for nonnegative sparse regression with self-dictionary | |
CN107770526B (en) | Hyperspectral atmospheric infrared remote sensing image lossless compression method based on quantitative ICA | |
CN109583380B (en) | Hyperspectral classification method based on attention-constrained non-negative matrix factorization | |
CN112131931B (en) | Deep forest hyperspectral image classification method and system based on attention mechanism | |
CN112633202A (en) | Hyperspectral image classification algorithm based on dual denoising combined multi-scale superpixel dimension reduction | |
CN111062428A (en) | Hyperspectral image clustering method, system and equipment | |
CN116051896A (en) | Hyperspectral image classification method of lightweight mixed tensor neural network | |
JP4981122B2 (en) | Suppressed vector quantization | |
CN114120020A (en) | Hyperspectral image inter-spectrum sequencing method based on key channel protection and spectral clustering | |
Shang et al. | Llava-prumerge: Adaptive token reduction for efficient large multimodal models | |
CN109089118B (en) | Hyperspectral atmospheric infrared remote sensing image compression method based on key information protection | |
CN106101732B (en) | The vector quantization scheme of Fast Compression bloom spectrum signal | |
CN114119787B (en) | Hyper-spectral image prediction compression method based on orthogonal matching pursuit | |
Qian et al. | 3D data compression of hyperspectral imagery using vector quantization with NDVI-based multiple codebooks | |
CN107945131B (en) | Distributed image reconstruction method for radio interference array | |
CA2054849C (en) | Speech parameter encoding method capable of transmitting a spectrum parameter at a reduced number of bits | |
Ülkü et al. | Lossy compressive sensing based on online dictionary learning | |
JP3228389B2 (en) | Gain shape vector quantizer | |
JP5244452B2 (en) | Document feature expression calculation apparatus and program | |
Vázquez-Padín et al. | Derivation of the asymptotic eigenvalue distribution for causal 2D-AR models under upscaling |
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 |