CN112417188B - Hyperspectral image classification method based on graph model - Google Patents

Hyperspectral image classification method based on graph model Download PDF

Info

Publication number
CN112417188B
CN112417188B CN202011436246.5A CN202011436246A CN112417188B CN 112417188 B CN112417188 B CN 112417188B CN 202011436246 A CN202011436246 A CN 202011436246A CN 112417188 B CN112417188 B CN 112417188B
Authority
CN
China
Prior art keywords
node
formula
matrix
value
equation
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202011436246.5A
Other languages
Chinese (zh)
Other versions
CN112417188A (en
Inventor
蒋俊正
黄炟鑫
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Guilin University of Electronic Technology
Original Assignee
Guilin University of Electronic Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Guilin University of Electronic Technology filed Critical Guilin University of Electronic Technology
Priority to CN202011436246.5A priority Critical patent/CN112417188B/en
Publication of CN112417188A publication Critical patent/CN112417188A/en
Application granted granted Critical
Publication of CN112417188B publication Critical patent/CN112417188B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
    • G06F16/55—Clustering; Classification
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
    • G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
    • G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
    • G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
    • G06F16/5866—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using information manually generated, e.g. tags, keywords, comments, manually generated location and time information
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/901—Indexing; Data structures therefor; Storage structures
    • G06F16/9024—Graphs; Linked lists
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
    • G06F18/24147—Distances to closest patterns, e.g. nearest neighbour classification

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • Databases & Information Systems (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Library & Information Science (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Computation (AREA)
  • Evolutionary Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Software Systems (AREA)
  • Image Analysis (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The invention discloses a hyperspectral image classification method based on a graph model, which is characterized by comprising the following steps of: 1) modeling a graph; 2) resolving an optimization problem; 3) decomposing a Hessian matrix; 4) an approximation of newton step size; 5) carrying out iterative solution; 6) distributed solving; 7) and solving the classification result. The method has low calculation complexity under large-scale data and can finish classification with higher precision.

Description

Hyperspectral image classification method based on graph model
Technical Field
The invention relates to the technical field of hyperspectral images, graph models and machine learning, in particular to a hyperspectral image classification method based on a graph model.
Background
The hyperspectral image data is a special image, is firstly applied to the satellite remote sensing technology, and is widely applied to the fields of military affairs, medicine, geology, agriculture and the like. The common image is subjected to broadband imaging during imaging, the imaging result hardly has spectral characteristics, and the related research is extremely dependent on the spatial resolution of the image. The hyperspectral image is the result of imaging of multiple narrow bands, and compared with a common image, the hyperspectral image carries abundant spectrum information. The research on the hyperspectral data is very extensive, and classification is an important item. The hyperspectral image classification aims at marking pixel points, and the marking result can well reflect the relevant characteristics of the target to be detected. However, the hyperspectral image classification has the problems of small known sample size, high data dimensionality and the like.
The diagram is a common data structure, and is commonly used in wireless sensor networks, social networks and other scenarios. In addition, graphs are also commonly used for dimensionality reduction of data with higher feature dimensions.
Machine learning is a general term for a type of task solution method, which mathematically models a related task and then uses a computer or other tool to solve the modeled optimization problem to obtain a result. Machine learning has been widely used in industry, traffic, IC design, etc. through decades of development.
Disclosure of Invention
The invention aims to provide a hyperspectral image classification method based on a graph model, aiming at the problems of large data size, high data dimensionality, small known sample size and the like in a hyperspectral image classification task. The method has low calculation complexity under large-scale data and can finish classification with higher precision.
The technical scheme for realizing the purpose of the invention is as follows:
a hyperspectral image classification method based on a graph model comprises the following steps:
1) graph modeling: suppose the hyperspectral image data is X ═ X1,x2,...,xm×n]∈Rm×n×dTotal of m × n pixel points, xi∈RdIs the feature vector of the ith pixel,
Figure GDA0003574567220000011
labels on pixels in the data are all from L ═ L 1,l2,...,lcAnd all the types of the C are replaced by C,
Figure GDA0003574567220000012
is a tag of a known moiety in X,
Figure GDA0003574567220000013
is an unknown tag, has a value of 0, and
Figure GDA0003574567220000014
feature vector x of different pixel points of hyperspectral image dataiThe graph G is constructed as { V, E, W }, where V is a node set of the graph, corresponding to each pixel in the data and connected by an edge, E is an edge set, the edge is used to describe similarity and adjacency relation between nodes, W is a weight matrix, and an internal element represents a similarity degree corresponding to two nodes and is defined as formula (1):
Figure GDA0003574567220000015
Figure GDA0003574567220000021
is a neighbor set of the node i;
2) the optimization problem is summarized as follows: the one-hot encoding matrix Y of the label Y is defined as shown in formula (2):
Figure GDA0003574567220000022
yiis the ith element in tag y. From Y, the classification problem of hyperspectral images can be summarized as shown in equation (3):
Figure GDA0003574567220000023
and (3) solving the formula (3) to obtain a one-hot coding matrix F for classification, wherein in the formula (3), alpha is a weight factor, and L isn＝I-D-1/2WD-1/2Is a normalized graph Laplace matrix, I is a unit matrix, yjAnd fjThe j-th columns of Y and F respectively,
Figure GDA0003574567220000024
is fjD is a degree matrix, defined as shown in formula (4):
Figure GDA0003574567220000025
will D-1/2WD-1/2Is marked as WnThe problem in the formula (3) is composed of c optimization problems, which are independent from each other and similar in solving process, and only one of the optimization problems is analyzed, so that f is equal to f jThe method comprises the following steps:
Figure GDA0003574567220000026
in the formula (5)
Figure GDA0003574567220000027
In order to match the terms of the search term,
Figure GDA0003574567220000028
is a regular term;
3) decomposition of the Hessian matrix: the problem in the formula (5) is solved by using a Newton method, and the information of the first order and the second order gradient of the problem is respectively shown in formulas (6) and (7):
Figure GDA0003574567220000029
Figure GDA00035745672200000210
thus obtaining the Hessian matrix H ═ alpha I + L of the objective functionnLet Wn＝Wnd+WnuWherein W isnd＝diag(Wn) From Ln＝I-WnThe method comprises the following steps:
A＝αI+(1+θ)(I-Wnd) (8)，
B＝Wnu+θ(I-Wnd) (9)，
wherein theta is a parameter for controlling decomposition, and the Hessian matrix can be decomposed into a combination of A and B as shown in formula (10):
H＝A-B (10)；
4) approximation of newton step size: when solving by the Newton method, the Newton step length of the k step is represented by the formula (10):
Figure GDA00035745672200000211
let P equal Wn-αI，Q＝(I-P)-1Q ≈ I + P by Taylor expansion, the Newton step size of equation (11) is approximated by equation (12):
Figure GDA0003574567220000031
5) and (3) iterative solution: equation (5) is solved according to equation (13) using the approximate newton step size in equation (12):
Figure GDA0003574567220000032
in equation (13), k in the upper right corner of the variable is the number of iterations, and f0＝0，f1When 1 is equal to
Figure GDA0003574567220000033
When the result is true, terminating the iteration;
6) distributed solution: the matching terms in the formula (5) are respectively expressed as
Figure GDA0003574567220000034
The first order gradient information at the ith node is shown in equation (14):
Figure GDA0003574567220000035
wijis WnThe element in the ith row and j column corresponds to the weight value between the node i and the node j on the graph, so that for the node i, the following elements exist:
6-1) with
Figure GDA0003574567220000036
The node in (1) communicates to obtain f kIn that
Figure GDA0003574567220000037
A value of (d) above;
6-2) calculation
Figure GDA0003574567220000038
Figure GDA0003574567220000039
Is ekValue at node i, AiiIs the ith element on the diagonal of A;
6-3) with
Figure GDA00035745672200000310
The node in (1) communicates to obtain e at
Figure GDA00035745672200000311
A value of (d) above;
6-4) calculation
Figure GDA00035745672200000312
Figure GDA00035745672200000313
Is ukThe value at node i;
6-5) with
Figure GDA00035745672200000314
The node in (1) communicates to obtain u is
Figure GDA00035745672200000315
A value of (d);
6-6) calculating the step size
Figure GDA00035745672200000316
Figure GDA00035745672200000317
Is the step length skThe value at node i;
6-7) calculation
Figure GDA00035745672200000318
To fkUpdating the value at the node i;
6-8) on all nodes
Figure GDA00035745672200000319
Are spliced into fk+1Judging the termination condition
Figure GDA00035745672200000320
Whether or not: if the condition is terminated
Figure GDA00035745672200000321
If it is true, fk+1Setting a corresponding column in the one-hot matrix, and starting to solve the next column of the one-hot matrix; if the condition is terminated
Figure GDA00035745672200000322
If not, f is usedk+1Performing the next iteration with the initial value;
7) solving a classification result: obtaining a one-hot coding matrix F through the step 6), and finally obtaining a classification result through a formula (15)
Figure GDA00035745672200000323
Figure GDA00035745672200000324
The technical scheme includes that a graph model is adopted to carry out dimensionality reduction on hyperspectral image data, then a learning task is subjected to problem regression, then a Hessian matrix is approximated through matrix decomposition, and the approximated Hessian matrix is used for solving through a quasi-Newton method.
The method has the advantages of low calculation complexity and high classification precision, and is more suitable for the large-scale hyperspectral image classification problem.
Drawings
FIG. 1 is a schematic diagram of iterative convergence in KSC data for implementing the method;
fig. 2 is a schematic diagram of iterative convergence in IndianPines data by the implementation method.
Detailed Description
The invention will be further described with reference to the following drawings and examples, but is not limited thereto.
Example (b):
a hyperspectral image classification method based on a graph model comprises the following steps:
1) graph modeling: suppose the hyperspectral image data is X ═ X1,x2,...,xm×n]∈Rm×n×dTotal of m × n pixel points, xi∈RdIs the feature vector of the ith pixel,
Figure GDA0003574567220000041
labels on pixels in the data are all from L ═ L1,l2,...,lcAnd all the types of the C are replaced by C,
Figure GDA0003574567220000042
is a tag of a known moiety in X,
Figure GDA0003574567220000043
is an unknown tag, has a value of 0, and
Figure GDA0003574567220000044
feature vector x of different pixel points of hyperspectral image dataiThe graph G is constructed as { V, E, W }, where V is a node set of the graph, corresponding to each pixel in the data and connected by an edge, E is an edge set, the edge is used to describe similarity and adjacency relation between nodes, W is a weight matrix, and an internal element represents a similarity degree corresponding to two nodes and is defined as formula (1):
Figure GDA0003574567220000045
Figure GDA0003574567220000046
is a neighbor set of the node i;
2) the optimization problem is summarized as follows: the one-hot encoding matrix Y of the label Y is defined as shown in formula (2):
Figure GDA0003574567220000047
yiIs the ith element in tag y. From Y, the classification problem of hyperspectral images can be summarized as shown in equation (3):
Figure GDA0003574567220000048
and (3) solving the formula (3) to obtain a one-hot coding matrix F for classification, wherein in the formula (3), alpha is a weight factor, and L isn＝I-D-1/2WD-1/2Is a normalized graph Laplace matrix, I is a unit matrix, yjAnd fjThe j-th columns of Y and F respectively,
Figure GDA0003574567220000049
is fjD is a degree matrix, defined as shown in formula (4):
Figure GDA00035745672200000410
will D-1/2WD-1/2Is marked as WnThe problem in the formula (3) is composed of c optimization problems, which are independent from each other and similar in solving process, and only one of the optimization problems is analyzed, so that f is equal to fjThe method comprises the following steps:
Figure GDA0003574567220000051
in the formula (5)
Figure GDA0003574567220000052
In order to match the terms of the search,
Figure GDA0003574567220000053
is a regular term;
3) decomposition of the Hessian matrix: the problem in the formula (5) is solved by using a Newton method, and the information of the first order and the second order gradient of the problem is respectively shown in formulas (6) and (7):
Figure GDA0003574567220000054
Figure GDA0003574567220000055
thus obtaining the Hessian matrix H ═ alpha I + L of the objective functionnLet Wn＝Wnd+WnuWherein W isnd＝diag(Wn) From Ln＝I-WnThe method comprises the following steps:
A＝αI+(1+θ)(I-Wnd) (8)，
B＝Wnu+θ(I-Wnd) (9)，
wherein theta is a parameter for controlling decomposition, and the Hessian matrix can be decomposed into a combination of A and B as shown in formula (10):
H＝A-B (10)；
4) approximation of newton step size: when the Newton method is used for solving, the Newton step length of the k step is represented by the formula (10):
Figure GDA0003574567220000056
let P equal Wn-αI，Q＝(I-P)-1Q ≈ I + P by Taylor expansion, the Newton step size of equation (11) is approximated by equation (12):
Figure GDA0003574567220000057
5) And (3) iterative solution: equation (5) is solved using the approximate newton step size in equation (12) according to equation (13):
Figure GDA0003574567220000058
in equation (13), k at the upper right corner of the variable is the number of iterations, and f0＝0，f1When being equal to 1
Figure GDA0003574567220000059
When the result is true, terminating the iteration;
6) distributed solution: the matching terms in the formula (5) are respectively expressed as
Figure GDA00035745672200000510
The first order gradient information at the ith node is shown in equation (14):
Figure GDA0003574567220000061
wijis WnThe element in the ith row and j column corresponds to the weight value between the node i and the node j on the graph, so that for the node i, the following elements exist:
6-1) with
Figure GDA0003574567220000062
The node in (1) communicates to obtain fkIn that
Figure GDA0003574567220000063
A value of (d) above;
6-2) calculation
Figure GDA0003574567220000064
Figure GDA0003574567220000065
Is ekValue at node i, AiiIs the ith element on the diagonal of A;
6-3) with
Figure GDA0003574567220000066
The node in (1) communicates to obtain e at
Figure GDA0003574567220000067
A value of (d) above;
6-4) calculation
Figure GDA0003574567220000068
Figure GDA0003574567220000069
Is ukThe value at node i;
6-5) with
Figure GDA00035745672200000610
The node in (1) communicates to obtain u is
Figure GDA00035745672200000611
A value of (d);
6-6) calculating the step size
Figure GDA00035745672200000612
Figure GDA00035745672200000613
Is a step size skThe value at node i;
6-7) calculation
Figure GDA00035745672200000614
To fkUpdating the value at the node i;
6-8) on all nodes
Figure GDA00035745672200000615
Are spliced into fk+1Judging the termination condition
Figure GDA00035745672200000616
Whether or not: if the condition is terminated
Figure GDA00035745672200000617
If it is true, fk+1Setting a corresponding column in the one-hot matrix, and starting to solve the next column of the one-hot matrix; if the condition is terminated
Figure GDA00035745672200000618
If not, use fk+1Performing the next iteration for the initial value;
7) solving a classification result: obtaining a one-hot coding matrix F by the step 6), and finallyObtaining the classification result through the formula (15)
Figure GDA00035745672200000619
Figure GDA00035745672200000620
Simulation example 1:
in the embodiment, KSC data is adopted for simulation, 176 frequency bands are arranged in the KSC data, 5211 pixel points are total after background removal, the pixel points can be divided into 13 classes and are a multi-class learning task, before simulation, tag values are sampled to generate data sets with different known tag numbers, for multi-item comparison, data sets with 5, 10, 15 and 20 known samples in each class are generated, table 1 shows that the classification performance comparison in the KSC data by adopting the method of the embodiment and the centralized method is performed, the OA parameters and the calculation time are included, and the simulation result of table 1 shows that in the KSC data, compared with the centralized method, the classification accuracy by adopting the method of the embodiment and the centralized method is completely consistent, the calculation speed by adopting the method of the embodiment is slightly higher than that by adopting the centralized method, as shown in figure 1, iterative convergence can be performed within limited times by adopting the method of the embodiment, it is demonstrated that the method of the present example produces better results and is less time consuming.
TABLE 1
Number of samples known 5 10 15 20
OA value Using the method of this example 66.75％ 71.76％ 73.87％ 75.03％
Using this example method time consuming(s) 7.476 8.181 8.057 8.153
Using centralized approach OA values 66.5％ 71.78％ 73.87％ 75.03％
Consuming time using a centralized approach(s) 35.72 34.97 35.08 35.55
Simulation example 2:
in the embodiment, IndanPines data is adopted for simulation, 200 frequency bands exist in the IndanPines data, 10249 pixel points are obtained after background removal, the pixel points can be divided into 16 classes and are also a multi-class learning task, as in simulation example 1, sampling produced datasets of only 5, 10, 15, 20 known samples for each class, and table 2 shows the comparison of classification performance in indianpins data using the method of the present example and the centralized method, which also includes OA parameters and the time consumption of calculation, the simulation results in table 2 show that, in Indian pines data, compared with the centralized method, the classification accuracy of the method is basically consistent with that of the centralized method, in the simulation example, the calculation speed of the method is far better than that of the method adopting a centralized method, and as shown in fig. 2, iterative convergence can be performed within a limited number of times by adopting the method of the example, which shows that the method of the example is more suitable for a hyperspectral image classification task under large-scale data.
TABLE 2
Number of samples known 5 10 15 20
OA value Using the method of this example 41.98％ 46.69％ 49.71％ 51.67％
Using this example method time consuming(s) 37.09 38.59 39.1 39.83
Using centralized approach OA values 42.02％ 46.7％ 49.72％ 51.68％
Consuming time using a centralized approach(s) 242.9 271 242.5 299.6
。

Claims (1)

1. A hyperspectral image classification method based on a graph model is characterized by comprising the following steps:
1) graph modeling: suppose the hyperspectral image data is X ═ X1,x2,...,xm×n]∈Rm×n×dTotal of m × n pixel points, xi∈RdIs the feature vector of the ith pixel,
Figure FDA0003574567210000011
labels on pixels in the data are all from L ═ L1,l2,...,lcAnd all the types of the C are replaced by C,
Figure FDA0003574567210000012
is a tag of a known moiety in X,
Figure FDA0003574567210000013
is an unknown tag, has a value of 0, and
Figure FDA0003574567210000014
feature vector x of different pixel points of hyperspectral image dataiAccording to the similar characteristics of the method, a graph G is constructed, wherein the graph G is { V, E, W }, V is a node set of the graph and corresponds to each pixel in the dataThe nodes are connected by edges, E is an edge set, the edges are used for describing similarity and adjacency relations between the nodes, W is a weight matrix, and internal elements represent the similarity degree of the corresponding two nodes and are defined as shown in a formula (1):
Figure FDA0003574567210000015
Figure FDA0003574567210000016
is a neighbor set of the node i;
2) the optimization problem is summarized as follows: the one-hot encoding matrix Y of the label Y is defined as shown in formula (2):
Figure FDA0003574567210000017
yiis the ith element in label y; from Y, the classification problem of hyperspectral images can be summarized as shown in equation (3):
Figure FDA0003574567210000018
and (3) solving the formula (3) to obtain a one-hot coding matrix F for classification, wherein in the formula (3), alpha is a weight factor, and L is n＝I-D-1/2WD-1/2Is a normalized graph Laplace matrix, I is a unit matrix, yjAnd fjThe j-th columns of Y and F respectively,
Figure FDA0003574567210000019
is fjD is a degree matrix, defined as shown in formula (4):
Figure FDA00035745672100000110
will D-1/2WD-1/2Is marked as WnThe problem in the formula (3) is composed of c optimization problems, which are independent from each other and similar in solving process, and only one of the optimization problems is analyzed, so that f is equal to fjThe method comprises the following steps:
Figure FDA00035745672100000111
in the formula (5)
Figure FDA00035745672100000112
In order to match the terms of the search,
Figure FDA00035745672100000113
is a regular term;
3) decomposition of the Hessian matrix: the problem in the formula (5) is solved by using a Newton method, and the information of the first order and the second order gradient of the problem is respectively shown in formulas (6) and (7):
Figure FDA00035745672100000114
Figure FDA0003574567210000021
thus obtaining the Hessian matrix H ═ alpha I + L of the objective functionnLet Wn＝Wnd+WnuWherein W isnd＝diag(Wn) From Ln＝I-WnThe method comprises the following steps:
A＝αI+(1+θ)(I-Wnd) (8)，
B＝Wnu+θ(I-Wnd) (9)，
wherein theta is a parameter for controlling decomposition, and the Hessian matrix can be decomposed into a combination of A and B as shown in formula (10):
H＝A-B (10)；
4) approximation of newton step size: when solving by the Newton method, the Newton step length of the k step is represented by the formula (10):
Figure FDA0003574567210000022
let P equal Wn-αI，Q＝(I-P)-1Q ≈ I + P by Taylor expansion, the Newton step size of equation (11) is approximated by equation (12):
Figure FDA0003574567210000023
5) and (3) iterative solution: equation (5) is solved according to equation (13) using the approximate newton step size in equation (12):
Figure FDA0003574567210000024
in equation (13), k in the upper right corner of the variable is the number of iterations, and f 0＝0，f1When 1 is equal to
Figure FDA0003574567210000025
When the result is true, terminating the iteration;
6) distributed solution: the matching terms in the formula (5) are respectively expressed as
Figure FDA0003574567210000026
The first order gradient information at the ith node is shown in equation (14):
Figure FDA0003574567210000027
wijis WnThe element in the ith row and j column corresponds to the weight value between the node i and the node j on the graph, so that for the node i, the following elements exist:
6-1) with
Figure FDA0003574567210000028
The node in (1) communicates to obtain fkIn that
Figure FDA0003574567210000029
A value of (d) above;
6-2) calculation
Figure FDA00035745672100000210
Figure FDA00035745672100000211
Is ekValue at node i, AiiIs the ith element on the diagonal of A;
6-3) with
Figure FDA00035745672100000212
The node in (1) communicates to obtain e at
Figure FDA00035745672100000213
A value of (d) above;
6-4) calculation
Figure FDA00035745672100000214
Figure FDA00035745672100000215
Is ukThe value at node i;
6-5) with
Figure FDA0003574567210000031
The node in (1) communicates to obtain u is
Figure FDA0003574567210000032
A value of (d);
6-6) calculating the step size
Figure FDA0003574567210000033
Figure FDA0003574567210000034
Is the step length skThe value at node i;
6-7) calculation
Figure FDA0003574567210000035
To fkUpdating the value at the node i;
6-8) will f on all nodesi k+1Are spliced into fk+1Judging the termination condition
Figure FDA0003574567210000036
Whether or not: if the condition is terminated
Figure FDA0003574567210000037
If it is true, fk+1Setting a corresponding column in the one-hot matrix, and starting to solve the next column of the one-hot matrix; if the condition is terminated
Figure FDA0003574567210000038
If not, f is usedk+1Performing the next iteration with the initial value;
7) solving a classification result: obtaining a one-hot coding matrix F through the step 6), and finally obtaining a classification result through a formula (15)
Figure FDA0003574567210000039
Figure FDA00035745672100000310
CN202011436246.5A 2020-12-10 2020-12-10 Hyperspectral image classification method based on graph model Active CN112417188B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011436246.5A CN112417188B (en) 2020-12-10 2020-12-10 Hyperspectral image classification method based on graph model

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011436246.5A CN112417188B (en) 2020-12-10 2020-12-10 Hyperspectral image classification method based on graph model

Publications (2)

Publication Number Publication Date
CN112417188A CN112417188A (en) 2021-02-26
CN112417188B true CN112417188B (en) 2022-05-24

Family

ID=74775010

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011436246.5A Active CN112417188B (en) 2020-12-10 2020-12-10 Hyperspectral image classification method based on graph model

Country Status (1)

Country Link
CN (1) CN112417188B (en)

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106934400A (en) * 2017-02-24 2017-07-07 西安电子科技大学 Based on the hyperspectral image band selection method from expression and local similarity protection
CN111860612A (en) * 2020-06-29 2020-10-30 西南电子技术研究所（中国电子科技集团公司第十研究所） Unsupervised hyperspectral image hidden low-rank projection learning feature extraction method

Family Cites Families (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090318815A1 (en) * 2008-05-23 2009-12-24 Michael Barnes Systems and methods for hyperspectral medical imaging
US8670620B2 (en) * 2011-08-10 2014-03-11 National Ict Australia Limited Decomposing hyperspectral or multispectral image data
CN102521601B (en) * 2011-11-17 2013-12-25 西安电子科技大学 Method for classifying hyperspectral images based on semi-supervised conditional random field
WO2015192117A1 (en) * 2014-06-14 2015-12-17 Magic Leap, Inc. Methods and systems for creating virtual and augmented reality
TWI624807B (en) * 2017-09-27 2018-05-21 Taichung Veterans General Hosital Iterative analysis of medical images
CN108470192B (en) * 2018-03-13 2022-04-19 广东工业大学 Hyperspectral classification method and device
CN108764276A (en) * 2018-04-12 2018-11-06 西北大学 A kind of robust weights multi-characters clusterl method automatically
CN110717354B (en) * 2018-07-11 2023-05-12 哈尔滨工业大学 Super-pixel classification method based on semi-supervised K-SVD and multi-scale sparse representation
CN110232317B (en) * 2019-05-05 2023-01-03 五邑大学 Hyper-spectral image classification method, device and medium based on super-pixel segmentation and two-stage classification strategy
CN110399909B (en) * 2019-07-08 2021-07-27 南京信息工程大学 Hyperspectral image classification method based on label constraint elastic network graph model
CN111242867B (en) * 2020-01-14 2023-06-13 桂林电子科技大学 Graph signal distributed on-line reconstruction method based on truncated taylor series approximation
CN111275201A (en) * 2020-01-21 2020-06-12 桂林电子科技大学 Sub-graph division based distributed implementation method for semi-supervised learning of graph

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106934400A (en) * 2017-02-24 2017-07-07 西安电子科技大学 Based on the hyperspectral image band selection method from expression and local similarity protection
CN111860612A (en) * 2020-06-29 2020-10-30 西南电子技术研究所（中国电子科技集团公司第十研究所） Unsupervised hyperspectral image hidden low-rank projection learning feature extraction method

Also Published As

Publication number Publication date
CN112417188A (en) 2021-02-26

Similar Documents

Publication Publication Date Title
Lei et al. Intelligent fault detection of high voltage line based on the Faster R-CNN
CN110717354B (en) Super-pixel classification method based on semi-supervised K-SVD and multi-scale sparse representation
Kang et al. Dual-path network-based hyperspectral image classification
CN108009559B (en) Hyperspectral data classification method based on space-spectrum combined information
Zass et al. Probabilistic graph and hypergraph matching
Montazer et al. An improved radial basis function neural network for object image retrieval
Zhu et al. Unsupervised classification in hyperspectral imagery with nonlocal total variation and primal-dual hybrid gradient algorithm
Jia et al. ROML: A robust feature correspondence approach for matching objects in a set of images
Qin et al. Cross-domain collaborative learning via cluster canonical correlation analysis and random walker for hyperspectral image classification
CN109190511B (en) Hyperspectral classification method based on local and structural constraint low-rank representation
CN112906756A (en) High-image classification method and system for cross-channel quantity transfer learning
CN110705636A (en) Image classification method based on multi-sample dictionary learning and local constraint coding
CN108470192B (en) Hyperspectral classification method and device
CN107273919B (en) Hyperspectral unsupervised classification method for constructing generic dictionary based on confidence
CN111340108A (en) High-dimensional data visualization method based on optimal transmission theory
CN116643246A (en) Deep clustering radar pulse signal sorting method based on inner product distance measurement
CN113705674B (en) Non-negative matrix factorization clustering method and device and readable storage medium
Zhao et al. Hyperspectral target detection method based on nonlocal self-similarity and rank-1 tensor
CN107578445B (en) Image discriminable region extraction method based on convolution characteristic spectrum
Qin et al. Fast blind hyperspectral unmixing based on graph laplacian
Huang et al. Hyperspectral image clustering via sparse dictionary‐based anchored regression
CN112417188B (en) Hyperspectral image classification method based on graph model
CN111062888B (en) Hyperspectral image denoising method based on multi-target low-rank sparsity and spatial-spectral total variation
CN116563649A (en) Tensor mapping network-based hyperspectral image lightweight classification method and device
CN108009586B (en) Capping concept decomposition method and image clustering 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