CN111897988A - A kind of hyperspectral remote sensing image classification method and system - Google Patents

Abstract

The invention relates to a hyperspectral remote sensing image classification method and system. The method includes acquiring a hyperspectral remote sensing image; determining an anchor point according to the pixel points of the hyperspectral remote sensing image; and constructing an adaptive two According to the self-adaptive bipartite graph, a semi-supervised learning method is used to construct a semi-supervised learning model of the bipartite graph; the hyperspectral remote sensing image to be classified is obtained; the semi-supervised learning model of the bipartite graph is used to The hyperspectral remote sensing images to be classified are classified. The method and system for classifying hyperspectral remote sensing images provided by the present invention can reduce computational complexity and improve classification accuracy.

Description

A kind of hyperspectral remote sensing image classification method and system

technical field

The invention relates to the fields of hyperspectral remote sensing and machine learning, in particular to a hyperspectral remote sensing image classification method and system.

Background technique

Hyperspectral remote sensing is hyperspectral resolution remote sensing. As a narrow-band imaging method, it can discover substances that cannot be detected in wide-band. It emerged in the 1980s and is an important earth observation technology. While hyperspectral remote sensing technology provides help to our real life, it also leads to a series of problems of information extraction and pattern recognition, which are mainly reflected in high-dimensional data processing and analysis. With the development of imaging spectroscopy technology, higher spectral resolution brings more spectral bands, and wider coverage brings a larger amount of data. Therefore, while providing rich spectral information, hyperspectral technology also poses new challenges for hyperspectral data processing.

Classification is an important area of hyperspectral data processing. Hyperspectral image classification is to use the spectral information and spatial information of ground objects, according to certain classification criteria, such as "clustering of objects", to assign a category label to each pixel in the image. Hyperspectral classification technology plays an important role in extracting thematic information and monitoring the dynamic changes of ground objects. It is widely used in the production of thematic maps, engineering exploration, traffic planning and management, environmental monitoring, land use and crop yield estimation.

The original remote sensing image classification method is the manual visual interpretation method, which has high requirements on the staff's geoscience knowledge and research and judgment experience, and the classification results are greatly affected by the staff's experience and knowledge reserve. Artificial vision is less efficient and consumes a lot of manpower and material resources. Due to the rapid increase in the amount of remote sensing data, the artificial visual method has been unable to meet the demand. The essence of the classification problem is pattern recognition. With the continuous development of computer technology, the use of machine learning methods can realize intelligent data analysis and obtain hidden relationships between data. Various classification algorithms based on machine learning have improved the classification effect of hyperspectral images to a certain extent.

Machine learning is a core subject in the field of artificial intelligence and pattern recognition. According to whether the classification information is used, it can be divided into supervised learning (Supervised Learning), Unsupervised Learning (Unsupervised Learning) and Semi-supervised Learning (SSL) . Supervised learning uses the class label information of the input data to perform iterative calculation with probability function, algebraic function or artificial neural network as the basis function model. Unsupervised learning does not use class label information and is mainly used in data clustering. Semi-supervised learning uses a small amount of labeled information and a large amount of unlabeled information to build a learning model. Due to the large amount of hyperspectral data, the labeling of samples requires a lot of manpower and material resources, so semi-supervised learning methods have important applications in hyperspectral image processing.

Graph-based semi-supervised learning methods have clear concepts and are easy to implement, and have received extensive attention in recent years. However, the traditional graph method has high computational complexity in the process of composition and matrix inversion, and cannot handle large-scale hyperspectral images. In addition, traditional graph methods generally input a fixed graph, and the quality of the graph directly affects the subsequent classification effect.

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide a hyperspectral remote sensing image classification method and system, which can reduce the computational complexity and improve the classification accuracy.

For achieving the above object, the present invention provides the following scheme:

A hyperspectral remote sensing image classification method, comprising:

Obtain hyperspectral remote sensing images;

An anchor point is determined according to the pixel points of the hyperspectral remote sensing image; the anchor point is a randomly selected pixel point of the hyperspectral remote sensing image; the number of the anchor points is less than the number of pixels in the hyperspectral remote sensing image;

According to the pixel points of the hyperspectral remote sensing image and the anchor points, the adaptive proximity assignment principle is adopted to construct an adaptive bipartite graph;

According to the adaptive bipartite graph, a semi-supervised learning method is used to construct a semi-supervised learning model of the bipartite graph; the semi-supervised learning model of the bipartite graph takes the adaptive bipartite graph as input, and uses the hyperspectral remote sensing The category of the image is output;

Obtain hyperspectral remote sensing images to be classified;

The hyperspectral remote sensing image to be classified is classified using the semi-supervised learning model of the bipartite graph.

Optionally, according to the pixel points of the hyperspectral remote sensing image and the anchor points, the adaptive proximity assignment principle is used to construct an adaptive bipartite graph, which specifically includes:

According to the pixel point of the hyperspectral remote sensing image and the anchor point, adopt the principle of adaptive proximity assignment to construct a similarity matrix;

An adaptive bipartite graph is constructed from the similarity matrix.

Optionally, according to the adaptive bipartite graph, a semi-supervised learning method is used to construct a semi-supervised learning model of the bipartite graph, which further includes:

The adaptive bipartite graph is optimized using a semi-supervised learning objective.

Optionally, according to the adaptive bipartite graph, a semi-supervised learning method is used to construct a semi-supervised learning model of the bipartite graph, specifically including:

确定目标函数；Z为相似性矩阵，U为新的相似矩阵，

为像素点的软标签矩阵，

为锚点的软标签矩阵，

表示所有像素点的标签矩阵，B为对角矩阵，α为规则化参数，L_S为拉普拉斯矩阵；Use the formula

Determine the objective function; Z is the similarity matrix, U is the new similarity matrix,

is the soft label matrix of pixels,

is the soft label matrix of anchor points,

Represents the label matrix of all pixels, B is the diagonal matrix, α is the regularization parameter, and L _S is the Laplace matrix;

The objective function is solved in an iterative optimization manner to obtain the labels of the pixel points; the labels are used to classify the hyperspectral remote sensing images.

A hyperspectral remote sensing image classification system, comprising:

Hyperspectral remote sensing image acquisition module, used to acquire hyperspectral remote sensing images;

The anchor point determination module is used for determining anchor points according to the pixel points of the hyperspectral remote sensing image; the anchor points are randomly selected pixel points of the hyperspectral remote sensing image; the number of the anchor points is less than the number of the hyperspectral remote sensing images. the number of pixels;

an adaptive bipartite graph building module, configured to construct an adaptive bipartite graph according to the pixel points and the anchor points of the hyperspectral remote sensing image, using the principle of adaptive proximity assignment;

The semi-supervised learning model building module of bipartite graph is used for constructing a semi-supervised learning model of bipartite graph by adopting semi-supervised learning method according to the adaptive bipartite graph; the semi-supervised learning model of the bipartite graph is based on the self- Adapt the bipartite map as input, and use the category of the hyperspectral remote sensing image as output;

The hyperspectral remote sensing image acquisition module to be classified is used to acquire the hyperspectral remote sensing image to be classified;

A classification module for classifying the hyperspectral remote sensing image to be classified by using the semi-supervised learning model of the bipartite graph.

Optionally, the adaptive bipartite graph building module specifically includes:

a similarity matrix construction unit, configured to construct a similarity matrix according to the pixel point of the hyperspectral remote sensing image and the anchor point, using an adaptive proximity assignment principle;

An adaptive bipartite graph construction unit, configured to construct an adaptive bipartite graph according to the similarity matrix.

Optionally, also include:

An optimization module for optimizing the adaptive bipartite graph using a semi-supervised learning objective.

Optionally, the semi-supervised learning model building module of the bipartite graph specifically includes:

确定目标函数；Z为相似性矩阵，U为新的相似矩阵，

为像素点的软标签矩阵，

为锚点的软标签矩阵，

表示所有像素点的标签矩阵，B为对角矩阵，α为规则化参数，L_S为拉普拉斯矩阵；Objective function determination unit for utilizing the formula

Determine the objective function; Z is the similarity matrix, U is the new similarity matrix,

is the soft label matrix of pixels,

is the soft label matrix of anchor points,

Represents the label matrix of all pixels, B is the diagonal matrix, α is the regularization parameter, and L _S is the Laplace matrix;

The label determination unit of the pixel point is used to solve the objective function in an iterative optimization manner to obtain the label of the pixel point; the label is used to classify the hyperspectral remote sensing image.

According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

The method and system for classifying hyperspectral remote sensing images provided by the present invention, by determining randomly selected pixel points of some hyperspectral remote sensing images as anchor points, and constructing an adaptive bipartite graph between the anchor points and the pixel points, reducing The parameters of the composition are reduced, and the computational complexity is reduced; according to the adaptive bipartite graph, a semi-supervised learning method is used to construct a semi-supervised learning model of the bipartite graph, and the constructed adaptive bipartite graph is continuously optimized. The quality of the graph is improved, which in turn improves the accuracy of the classification.

Description of drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings required in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some of the present invention. In the embodiments, for those of ordinary skill in the art, other drawings can also be obtained according to these drawings without creative labor.

1 is a schematic flowchart of a method for classifying hyperspectral remote sensing images provided by the present invention;

FIG. 2 is a schematic structural diagram of a hyperspectral remote sensing image classification system provided by the present invention.

Detailed ways

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

The purpose of the present invention is to provide a hyperspectral remote sensing image classification method and system, which can reduce the computational complexity and improve the classification accuracy.

In order to make the above objects, features and advantages of the present invention more clearly understood, the present invention will be described in further detail below with reference to the accompanying drawings and specific embodiments.

1 is a schematic flowchart of a method for classifying hyperspectral remote sensing images provided by the present invention. As shown in FIG. 1 , a method for classifying hyperspectral remote sensing images provided by the present invention includes:

n表示像元的个数，d表示每个像元的波段数。S101, acquiring a hyperspectral remote sensing image. The pixels of the hyperspectral remote sensing image are represented by a matrix as

n represents the number of pixels, and d represents the number of bands per pixel.

S102: Determine anchor points according to pixels of the hyperspectral remote sensing image; the anchor points are randomly selected pixels of the hyperspectral remote sensing image; the number of anchor points is less than the number of pixels in the hyperspectral remote sensing image. That is, m anchor points are selected from the original n pixels, and the anchor points are represented by a matrix as

S103 , according to the pixel points of the hyperspectral remote sensing image and the anchor points, adopt the principle of adaptive proximity assignment to construct an adaptive bipartite graph. The principle of adaptive neighbor assignment is that the closer the distance between two points, the greater the probability of belonging to the same class.

S103 specifically includes:

According to the pixel points and the anchor points of the hyperspectral remote sensing image, the similarity matrix Z is constructed by adopting the principle of adaptive proximity assignment.

Z is:

为相似矩阵Z中的第i行，γ为规则化参数。令

是一个向量，它的第j个元素为e_ij，因此，(1)式写成如下向量形式：Among them, z _ij is the i-th row and j-th column element in the similarity matrix Z,

is the ith row in the similarity matrix Z, and γ is the regularization parameter. make

is a vector whose j-th element is e _ij , therefore, equation (1) is written in the following vector form:

The Lagrangian function corresponding to formula (2) is:

where η and β _i ≥ 0 are Lagrange multipliers.

The optimal solution of _zi should satisfy the derivative of formula (3) with respect to _zi equal to 0, namely:

For the jth element in _zi , there are:

According to the KKT condition, zi _ij β _ij = 0, which can be obtained from formula (5):

因此，z_ij的解为：in

Therefore, the solution of z _ij is:

An adaptive bipartite graph is constructed from the similarity matrix.

The adaptive bipartite graph is:

S104, according to the adaptive bipartite graph, adopt a semi-supervised learning method to construct a semi-supervised learning model of the bipartite graph; the semi-supervised learning model of the bipartite graph takes the adaptive bipartite graph as an input, and uses the high The category of the spectral remote sensing image is output;

Also included before S104:

The adaptive bipartite graph is optimized using a semi-supervised learning objective.

或者

如下：The goal is to learn a new similarity matrix

or

as follows:

The closer the optimized similarity matrix S is to the given nearest neighbor matrix W, the better, so the following problems need to be solved:

According to the special structure of S and W in formula (8) and formula (9), formula (10) can be converted into:

S104 specifically includes:

确定目标函数；Z为相似性矩阵，U为新的相似矩阵，

为像素点的软标签矩阵，

为锚点的软标签矩阵，

表示所有像素点的标签矩阵，B为对角矩阵，α为规则化参数，L_S为拉普拉斯矩阵。Use the formula

Determine the objective function; Z is the similarity matrix, U is the new similarity matrix,

is the soft label matrix of pixels,

is the soft label matrix of anchor points,

Represents the label matrix of all pixels, B is the diagonal matrix, α is the regularization parameter, and L _S is the Laplacian matrix.

The objective function is solved in an iterative optimization manner to obtain the labels of the pixel points; the labels are used to classify the hyperspectral remote sensing images.

Among them, the specific determination process of the objective function is as follows:

Considering the optimal quality of graphs in graph-based semi-supervised learning, the following model is proposed:

分别表示原始数据和锚点的软标签矩阵。

表示所有数据(原始数据和锚点)的标签矩阵。由于训练样本是从原始数据中选出的一部分数据点，因此，Y可以写为

其中

B是对角矩阵，对角线上的第i个元素是规则化参数β_i。B也可以写为

α是规则化参数。in,

Soft label matrices representing raw data and anchors, respectively.

A matrix of labels representing all data (raw data and anchors). Since the training samples are a subset of data points selected from the original data, Y can be written as

in

B is a diagonal matrix, and the ith element on the diagonal is the regularization parameter β _i . B can also be written as

α is the regularization parameter.

是度矩阵，第i个对角线上的元素为d_i＝∑_js_ij。D_S也可以写为

其中

是一个对角矩阵，对角线上的元素为U的行和，

也是对角矩阵，对角线上的元素为U的列和，

由于公式(12)中的约束条件为

可以得到D_r＝I_n，其中

是单位矩阵。因此，

然后，L_S可采用如下方式进行标准化：In graph theory, L _S = D _S -S is the Laplace matrix,

is a degree matrix, and the element on the ith diagonal is di ₌ ∑ _j s _ij . D _S can also be written as

in

is a diagonal matrix, the elements on the diagonal are the row sums of U,

is also a diagonal matrix, the elements on the diagonal are the column sums of U,

Since the constraints in Equation (12) are

It can be obtained that D _r =In , _where

is the identity matrix. therefore,

Then, L _S can be normalized as follows:

是单位矩阵。in,

is the identity matrix.

Therefore, the corresponding final objective function after normalizing the Laplacian matrix is:

The process of solving formula (14) by iterative optimization is as follows:

When F, G are fixed, formula (14) is equivalent to:

According to the basic properties of the normalized Laplace matrix, the following relationship can be obtained:

where f _i is the ith row of F, d _i =∑ _j s _ij , g _j is the ith row of G, d _j =∑ _i s _ji .

According to the special structure of S in Equation (9), Equation (16) is equivalent to:

Therefore, equation (15) can also be written as:

v_i、u_i、z_i表示向量，他们中的第j个元素分别为v_ij、u_ij和z_ij。因此，对于每一个i，公式(18)可以写成如下的向量形式：Since formula (18) is independent for different i, the objective function corresponding to each i can be solved. make

v _i , _ui , and _zi represent vectors, and the jth elements in them are v _ij , u _ij and z _ij , respectively. Therefore, for each i, equation (18) can be written in vector form as follows:

Formula (19) has the same form as formula (2), and can be solved by the same method.

When U is fixed, formula (14) is equivalent to:

公式(20)可以确定为：make

Formula (20) can be determined as:

By derivation of J(Q), the following formula can be obtained:

Therefore, the final solution is:

其中

式(23)等价于：B _α can be written as

in

Equation (23) is equivalent to:

L₂₂＝B_αm+I_m，采用如下分块矩阵求逆公式求解公式(24)中的第一项：Let L ₁₁ =In ₊ B _αn ,

L ₂₂ =B _αm +I _m , use the following block matrix inversion formula to solve the first term in formula (24):

由于

求

的计算复杂度为O(n³)，对于大规模高光谱数据，计算量太大，因此，采用如下Woodbury矩阵(A-UCV)^-1＝A^-1+A^-1U(C^-1-VA^-1U)^-1VA^-1求解大规模矩阵C₁求逆问题，将计算复杂度降低到O(nm²)。in,

because

beg

The computational complexity is O(n ³ ), which is too large for large-scale hyperspectral data. Therefore, the following Woodbury matrix (A-UCV) ^-1 =A ^-1 +A ^-1 U(C ^-1 - VA ^-1 U) ^-1 VA ^-1 solves the large-scale matrix C ₁ inversion problem, reducing the computational complexity to O(nm ² ).

Based on the above derivation, we can get the final soft label matrix as:

According to the soft label matrix F, the label of the data point x _i is obtained as:

The computational complexity of the entire model is O(ndmt+nm ² ), while the computational complexity of traditional graph-based semi-supervised learning models is O(n ² d+n ³ ). Among them, n, m, d and t are the number of samples, the number of anchor points, the dimension and the number of iterations, respectively. Therefore, for processing large-scale hyperspectral data, classification can be performed quickly and accurately.

S105, obtaining a hyperspectral remote sensing image to be classified.

S106, using the semi-supervised learning model of the bipartite graph to classify the hyperspectral remote sensing image to be classified.

FIG. 2 is a schematic structural diagram of a hyperspectral remote sensing image classification system provided by the present invention. As shown in FIG. 2, a hyperspectral remote sensing image classification system provided by the present invention includes: a hyperspectral remote sensing image acquisition module 201, an anchor A point determination module 202 , an adaptive bipartite graph building module 203 , a semi-supervised learning model building module for bipartite graphs 204 , a hyperspectral remote sensing image acquisition module 205 to be classified, and a classification module 206 .

The hyperspectral remote sensing image acquisition module 201 is used to acquire hyperspectral remote sensing images.

The anchor point determination module 202 is configured to determine an anchor point according to the pixel points of the hyperspectral remote sensing image; the anchor point is a randomly selected pixel point of the hyperspectral remote sensing image; the number of the anchor points is less than that in the hyperspectral remote sensing image. the number of pixels.

The adaptive bipartite graph construction module 203 is configured to construct an adaptive bipartite graph according to the pixel points and the anchor points of the hyperspectral remote sensing image, using the principle of adaptive proximity assignment.

The semi-supervised learning model building module 204 of the bipartite graph is used for constructing a semi-supervised learning model of the bipartite graph according to the self-adaptive bipartite graph using a semi-supervised learning method; the semi-supervised learning model of the bipartite graph is based on the self- The adaptive bipartite map is used as input, and the category of the hyperspectral remote sensing image is used as output.

The hyperspectral remote sensing image acquisition module 205 to be classified is configured to acquire the hyperspectral remote sensing image to be classified.

The classification module 206 is used for classifying the hyperspectral remote sensing image to be classified by using the semi-supervised learning model of the bipartite graph.

The adaptive bipartite graph building module 203 specifically includes: a similarity matrix building unit and an adaptive bipartite graph building unit.

The similarity matrix construction unit is used to construct a similarity matrix according to the pixel points of the hyperspectral remote sensing image and the anchor points, using the principle of adaptive proximity assignment.

The adaptive bipartite graph construction unit is configured to construct an adaptive bipartite graph according to the similarity matrix.

The hyperspectral remote sensing image classification system provided by the present invention further comprises: an optimization module.

An optimization module is used to optimize the adaptive bipartite graph using a semi-supervised learning objective.

The semi-supervised learning model building module 204 of the bipartite graph specifically includes: an objective function determination unit and a pixel label determination unit.

确定目标函数；Z为相似性矩阵，U为新的相似矩阵，

为像素点的软标签矩阵，

为锚点的软标签矩阵，

表示所有像素点的标签矩阵，B为对角矩阵，α为规则化参数，L_S为拉普拉斯矩阵。The objective function determination unit is used to utilize the formula

Determine the objective function; Z is the similarity matrix, U is the new similarity matrix,

is the soft label matrix of pixels,

is the soft label matrix of anchor points,

Represents the label matrix of all pixels, B is the diagonal matrix, α is the regularization parameter, and L _S is the Laplacian matrix.

The label determination unit of the pixel point is used to solve the objective function in an iterative optimization manner to obtain the label of the pixel point; the label is used to classify the hyperspectral remote sensing image.

The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same and similar parts between the various embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant part can be referred to the description of the method.

In this paper, specific examples are used to illustrate the principles and implementations of the present invention. The descriptions of the above embodiments are only used to help understand the methods and core ideas of the present invention; meanwhile, for those skilled in the art, according to the present invention There will be changes in the specific implementation and application scope. In conclusion, the contents of this specification should not be construed as limiting the present invention.

Claims (8)

1. a hyperspectral remote sensing image classification method, is characterized in that, comprises: Obtain hyperspectral remote sensing images; An anchor point is determined according to the pixel points of the hyperspectral remote sensing image; the anchor point is a randomly selected pixel point of the hyperspectral remote sensing image; the number of the anchor points is less than the number of pixels in the hyperspectral remote sensing image; According to the pixel points of the hyperspectral remote sensing image and the anchor points, the adaptive proximity assignment principle is adopted to construct an adaptive bipartite graph; According to the adaptive bipartite graph, a semi-supervised learning method is used to construct a semi-supervised learning model of the bipartite graph; the semi-supervised learning model of the bipartite graph takes the adaptive bipartite graph as input, and uses the hyperspectral remote sensing The category of the image is output; Obtain hyperspectral remote sensing images to be classified; The hyperspectral remote sensing image to be classified is classified using the semi-supervised learning model of the bipartite graph. 2. A kind of hyperspectral remote sensing image classification method according to claim 1, it is characterized in that, described according to the pixel point of described hyperspectral remote sensing image and described anchor point, adopt adaptive proximity assignment principle, construct adaptive Two pictures, including: According to the pixel point of the hyperspectral remote sensing image and the anchor point, adopt the principle of adaptive proximity assignment to construct a similarity matrix; An adaptive bipartite graph is constructed from the similarity matrix. 3. A kind of hyperspectral remote sensing image classification method according to claim 1, is characterized in that, described according to described adaptive bipartite graph, adopts semi-supervised learning method, constructs the semi-supervised learning model of bipartite graph, before. Also includes: The adaptive bipartite graph is optimized using a semi-supervised learning objective. 4. a kind of hyperspectral remote sensing image classification method according to claim 1, is characterized in that, described according to described adaptive bipartite graph, adopts semi-supervised learning method, constructs the semi-supervised learning model of bipartite graph, specifically include: Use the formula

Determine the objective function; Z is the similarity matrix, U is the new similarity matrix,

is the soft label matrix of pixels,

is the soft label matrix of anchor points,

Represents the label matrix of all pixels, B is the diagonal matrix, α is the regularization parameter, and L _S is the Laplace matrix; The objective function is solved in an iterative optimization manner to obtain the labels of the pixel points; the labels are used to classify the hyperspectral remote sensing images. 5. A hyperspectral remote sensing image classification system, characterized in that, comprising: Hyperspectral remote sensing image acquisition module, used to acquire hyperspectral remote sensing images; The anchor point determination module is used for determining anchor points according to the pixel points of the hyperspectral remote sensing image; the anchor points are randomly selected pixel points of the hyperspectral remote sensing image; the number of the anchor points is less than the number of the hyperspectral remote sensing images. the number of pixels; an adaptive bipartite graph building module, configured to construct an adaptive bipartite graph according to the pixel points and the anchor points of the hyperspectral remote sensing image, using the principle of adaptive proximity assignment; The semi-supervised learning model building module of bipartite graph is used for constructing a semi-supervised learning model of bipartite graph by adopting semi-supervised learning method according to the adaptive bipartite graph; the semi-supervised learning model of the bipartite graph is based on the self- Adapt the bipartite map as input, and use the category of the hyperspectral remote sensing image as output; The hyperspectral remote sensing image acquisition module to be classified is used to acquire the hyperspectral remote sensing image to be classified; A classification module for classifying the hyperspectral remote sensing image to be classified by using the semi-supervised learning model of the bipartite graph. 6. A kind of hyperspectral remote sensing image classification system according to claim 5, is characterized in that, described adaptive bipartite graph building module specifically comprises: a similarity matrix construction unit, configured to construct a similarity matrix according to the pixel point of the hyperspectral remote sensing image and the anchor point, using an adaptive proximity assignment principle; An adaptive bipartite graph construction unit, configured to construct an adaptive bipartite graph according to the similarity matrix. 7. A kind of hyperspectral remote sensing image classification system according to claim 5, is characterized in that, also comprises: An optimization module for optimizing the adaptive bipartite graph using a semi-supervised learning objective. 8. a kind of hyperspectral remote sensing image classification system according to claim 5, is characterized in that, the semi-supervised learning model building module of described bipartite graph specifically comprises: Objective function determination unit for utilizing the formula

Determine the objective function; Z is the similarity matrix, U is the new similarity matrix,

is the soft label matrix of pixels,

is the soft label matrix of anchor points,

Represents the label matrix of all pixels, B is the diagonal matrix, α is the regularization parameter, and L _S is the Laplace matrix; The label determination unit of the pixel point is used to solve the objective function in an iterative optimization manner to obtain the label of the pixel point; the label is used to classify the hyperspectral remote sensing image.

Images