CN112036235A - Hyperspectral image target detection method and system - Google Patents
Hyperspectral image target detection method and system Download PDFInfo
- Publication number
- CN112036235A CN112036235A CN202010692849.5A CN202010692849A CN112036235A CN 112036235 A CN112036235 A CN 112036235A CN 202010692849 A CN202010692849 A CN 202010692849A CN 112036235 A CN112036235 A CN 112036235A
- Authority
- CN
- China
- Prior art keywords
- matrix
- end member
- hyperspectral image
- projection
- abundance
- 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.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/30—Noise filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/194—Terrestrial scenes using hyperspectral data, i.e. more or other wavelengths than RGB
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Astronomy & Astrophysics (AREA)
- Remote Sensing (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention provides a hyperspectral image target detection method and a hyperspectral image target detection system, which comprise the following steps: performing dimensionality reduction on the hyperspectral image, and then projecting hyperspectral image pixels by using an end member matrix and an abundance matrix to obtain hyperspectral image end member projection space information quantity characteristics; iteratively solving the objective function based on the hyperspectral image end member projection space information quantity characteristic quantity to obtain an end member matrix and an abundance matrix; determining the ground feature composition of each pixel based on the end member matrix and the abundance matrix; the target function introduces the end member projection space information quantity into an MVC-NMF method as constraint, and combines the minimum end member volume constraint of a convex geometric model, so that the space formed by the real end members has larger statistical information quantity, and the target function iteration process is limited in a hyperplane formed by the real end members through the constraint of the information quantity, thereby reducing the influence of interference information such as noise, external points and the like on the end member extraction, extracting a real target, and realizing the accurate detection of the sub-pixel level image of the hyperspectral image ground object.
Description
Technical Field
The invention relates to the technical field of hyperspectral remote sensing image processing, in particular to a hyperspectral image target detection method and a hyperspectral image target detection system.
Background
The hyperspectral remote sensing is used for simultaneously acquiring two-dimensional space information and spectral feature information of ground objects, the atlas is an important feature of a hyperspectral image, images of all wave bands are overlapped to form a hyperspectral image data cube, each pixel is a continuous spectral curve, and points on the curve are radiation values on the images of all the wave bands. Therefore, in the hyperspectral image processing, the pixel and ground object analysis of the hyperspectral image is mainly carried out from two dimensions of an image space and a spectrum space.
Because the imaging ground features in the nature are complex and various, the hyperspectral image is limited by the resolution of a sensor, the area corresponding to each pixel often contains different types of ground features which have different spectral response characteristics, and the mixed pixels generally exist in the hyperspectral remote sensing image. When mixed pixels exist in the hyperspectral image, the conventional classification detection method is not strict in forcibly classifying the mixed pixels into a certain class of targets, the conventional method is difficult to meet the requirement of accurate quantitative analysis of the hyperspectral image, and great challenges are brought to the fine classification detection of the hyperspectral image.
The solution to the problem of hyperspectral end member extraction of mixed pixels is to decompose the mixed pixels to the level of sub-pixels, a process also known as mixed pixel analysis. The mixed pixel analysis decomposes the image to sub-pixel level, decomposes different ground objects contained in the mixed pixel by using methods of geometry, statistics, modeling and the like, each type of ground object is called an end member, and the composition, namely the occupied proportion, of each type of ground object is called abundance. The mixed pixel analysis process mainly comprises several links such as Feature Extraction (Feature Extraction), Endmember Extraction (endmemer Extraction), and abundance unmixing (unmixing) based on a mixed model. By utilizing a hyperspectral mixed pixel analysis technology, the proportion of various ground objects in the mixed pixel can be obtained through the abundance unmixing of the end members, the composition of various ground objects in the pixel is determined, the possibility of image misclassification can be reduced, and the precision of hyperspectral image ground object detection is improved.
In recent years, a newly-developed non-negative matrix decomposition blind signal separation method is applied to mixed pixel analysis of a hyperspectral remote sensing image, and becomes an important branch of a hyperspectral image target detection research direction. Noise or abnormal points in a hyperspectral image are also called as outliers and are points with large numerical difference with other values in the image, and currently, a non-negative matrix decomposition mixed pixel analysis image target detection method based on end member constraint, such as a non-negative matrix decomposition method based on end member simplex volume, end member sparsity, end member smooth transition and other constraints, rarely considers the influence of the outliers and ground objects with small probability distribution, and often extracts the noise and the abnormal points as end members or causes target detection errors due to the influence of the noise and the abnormal points. Aiming at the problem that when the existing common end member extraction method is used for processing data with exterior points, the data is easily influenced by the exterior points to extract wrong end members or directly extract abnormal pixel elements as the end members, the method achieves the purposes of robustness to noise and abnormal points and accurate energy efficiency detection of targets in the hyperspectral images by seeking effective constraint.
Disclosure of Invention
In order to solve the problems of noise and abnormal point robustness existing in the existing image target detection technology based on non-negative matrix factorization, the invention provides a hyperspectral image target detection method, which comprises the following steps:
performing dimensionality reduction on the hyperspectral image, and initializing an end member matrix and an abundance matrix;
performing projection processing on the hyperspectral image pixels subjected to dimensionality reduction by using the initialized end member matrix and the abundance matrix to obtain hyperspectral image end member projection space information quantity characteristics;
iteratively solving a pre-constructed objective function based on the hyperspectral image end member projection space information quantity characteristic quantity to obtain an end member matrix and an abundance matrix;
determining the ground feature composition of each pixel based on the end member matrix and the abundance matrix;
and constructing the objective function based on the projection information quantity of the spatial information statistical model and the minimum end member volume constraint based on the convex surface geometric model.
Preferably, the performing the dimensionality reduction on the hyperspectral image includes:
setting the number of end elements as p;
and performing noise whitening processing on the hyperspectral image by using the positive and negative transformation of the minimum noise separation transformation MNF, and performing inverse transformation on the first p-1 wave bands to obtain a noise whitened image serving as the hyperspectral image R after dimension reduction.
Preferably, the initializing the end member matrix and the abundance matrix includes:
and randomly selecting p points in the hyperspectral image R as initial end members, and setting the initial value of the initial abundance matrix to be 0.
Preferably, the performing projection processing on the hyperspectral image pixel subjected to the dimensionality reduction by using the initialized end member matrix and the abundance matrix to obtain the hyperspectral image end member projection spatial information quantity characteristic includes:
projecting all pixels in the hyperspectral image R to an end member matrix to obtain a projection matrix;
and obtaining the projection space information quantity characteristic of the end member of the hyperspectral image based on the sum of all the non-negative characteristic values in the projection matrix.
Preferably, the projection matrix is calculated as follows:
M=(I-PE)R
where M is the projection matrix, PEA projection operator of the end member matrix E;
the projection operator PEIs calculated as follows:
PE=E(ETE)-1ET。
preferably, the calculation formula of the hyperspectral image end-member projection spatial information quantity features is as follows:
in the formula, JIC(E) Projecting spatial information quantity characteristics for the hyperspectral image end members, wherein M is a projection matrix; sigmaMIs the covariance matrix of the projection matrix M, alphaiAs a covariance matrix sigmaMThe ith non-negative eigenvalue of (c).
Preferably, the objective function is:
in the formula (I), the compound is shown in the specification,is a model error value, λ and η are regularization parameters for balancing the model error value, end-member simplex volume constraint, and end-member projection spatial information, respectively, JVC(E) Is a minimum end member volume constraint function, E is an end member matrix of the hyperspectral image R, C is an abundance matrix, p is the number of end members,is a p-dimensional row vector;
the minimum end member volume constraint function is as follows:
preferably, the iteratively solving the pre-constructed objective function to obtain the end member matrix and the abundance matrix includes:
Extracting an objective function optimization iteration rule based on the information content and the volume constraint non-negative matrix decomposition end member, and setting an iteration calculation termination condition;
and setting a sigma value, carrying out iterative solution on the objective function, and ending the iteration when an iteration termination condition is met.
Preferably: the objective function optimization iteration rule is as follows:
in the formula, Ei+1Is the i +1 th end member, C in the end member matrixi+1Is the abundance value, alpha, corresponding to the i +1 th end memberiAnd betaiThe iteration step sizes of the end member matrix and the abundance matrix are respectively.
Preferably, the end member matrix and the abundance matrix are obtained by iteratively solving the pre-constructed objective functionBefore still including: in optimising iteration rules by an objective function
Preferably, the determining the terrestrial composition of each pixel element based on the end-member matrix and the abundance matrix comprises:
carrying out spectrum matching on the spectrum corresponding to the end member matrix and a preset target spectrum to obtain a target object corresponding to each end member matrix;
and determining the composition of the ground objects included in the pixel elements based on the abundance matrix and the target objects corresponding to the end member matrixes.
Based on the same invention concept, the invention also provides a hyperspectral image target detection system, which comprises:
the initialization module is used for carrying out dimensionality reduction on the hyperspectral image and initializing an end member matrix and an abundance matrix;
the projection module is used for carrying out projection processing on the hyperspectral image pixels subjected to the dimension reduction processing by utilizing the initialized end member matrix and the abundance matrix to obtain hyperspectral image end member projection space information quantity characteristics;
the function solving module is used for iteratively solving a pre-constructed objective function based on the hyperspectral image end member projection space information quantity characteristic quantity to obtain an end member matrix and an abundance matrix;
the ground object determination module is used for determining the ground object composition of each pixel based on the end member matrix and the abundance matrix;
and constructing the objective function based on the projection information quantity of the spatial information statistical model and the minimum end member volume constraint based on the convex surface geometric model.
Compared with the prior art, the invention has the beneficial effects that:
the invention provides a hyperspectral image target detection method and a hyperspectral image target detection system, which comprise the following steps: performing dimensionality reduction on the hyperspectral image, and initializing an end member matrix and an abundance matrix; performing projection processing on the hyperspectral image pixels subjected to dimensionality reduction by using the initialized end member matrix and the abundance matrix to obtain hyperspectral image end member projection space information quantity characteristics; iteratively solving a pre-constructed objective function based on the hyperspectral image end member projection space information quantity characteristic quantity to obtain an end member matrix and an abundance matrix; determining the ground feature composition of each pixel based on the end member matrix and the abundance matrix; the target function is constructed based on the projection information quantity of a spatial information statistical model and the minimum end member volume constraint based on a convex geometric model, the end member projection spatial information quantity is used as the constraint and introduced into an MVC-NMF method, so that a space formed by the real end members has larger statistical information quantity, the target function iteration process is limited in a hyperplane formed by the real end members through the constraint of the information quantity, the influence of interference information such as noise, external points and the like on the end member extraction is reduced, the real target is extracted, and the accurate detection of the sub-pixel level image of the hyperspectral image ground object is realized.
Drawings
FIG. 1 is a schematic diagram of a hyperspectral image target detection method according to the invention;
FIG. 2 is a flow chart of a hyperspectral image target detection method according to an embodiment of the invention;
FIG. 3 is a diagram of a hyperspectral image target detection system according to the present invention.
Detailed Description
Noise or abnormal points in a hyperspectral image are also called as outliers and are points with large numerical difference with other values in the image, and at present, a non-negative matrix decomposition mixed pixel analysis end member extraction method based on end member constraint, such as a non-negative matrix decomposition method based on end member simplex volume, end member sparsity, end member smooth transition and other constraints, rarely considers the influence of the outliers and ground objects with small probability distribution, and often extracts the noise and the abnormal points as end members or causes detection errors due to the influence of the noise and the abnormal points. The invention discloses a hyperspectral image target detection method and system, and belongs to the technical field of hyperspectral remote sensing image processing, and aims to solve the problem that when a current common end member extraction method is used for processing data with exterior points, the current common end member extraction method is easily influenced by the exterior points to extract wrong end members or directly extract abnormal pixels as end members; the key point of the technical scheme is that the method comprises the following steps: optimizing a hyperspectral image end member projection space information quantity characteristic calculation method for hyperspectral image end member extraction optimization iteration; an end member extraction method for fusing projection information quantity based on a spatial information statistical model and minimum end member volume constraint based on a convex geometric model and using the projection information quantity as one constraint for non-negative matrix decomposition is further provided; further optimizing a non-negative matrix factorization end member extraction optimization iteration method based on information quantity and volume constraint; and fourthly, performing spectrum matching on the extracted end member spectrum and the target spectrum by using a spectrum matching method to detect the specified target. According to the method, effective constraint is sought, the purposes of noise and abnormal point robustness and accurate detection of ground objects in the hyperspectral image are achieved, and meanwhile, an effective hyperspectral image target detection solution under the conditions that the hyperspectral image has noise and the abnormal point robustness is obtained through strict formula deduction calculation.
For a better understanding of the present invention, reference is made to the following description taken in conjunction with the accompanying drawings and examples
Example 1
As shown in fig. 1, the present invention provides a hyperspectral image target detection method, which includes the following technical features:
s1, performing dimensionality reduction on the hyperspectral image, and initializing an end member matrix and an abundance matrix;
s2, performing projection processing on the hyperspectral image pixels subjected to the dimensionality reduction by using the initialized end member matrix and the abundance matrix to obtain hyperspectral image end member projection spatial information quantity characteristics;
s3, iteratively solving a pre-constructed objective function based on the hyperspectral image end member projection space information quantity characteristic quantity to obtain an end member matrix and an abundance matrix;
s4, determining the ground feature composition of each pixel based on the end member matrix and the abundance matrix;
and constructing the objective function based on the projection information quantity of the spatial information statistical model and the minimum end member volume constraint based on the convex surface geometric model.
The method for detecting the target of the hyperspectral image requires higher noise and abnormal point robustness, and finishes the design and the compilation of software on the basis of the existing minimum Volume constrained non-negative Matrix factorization (MVC-NMF) method research, and realizes the accurate target detection of the hyperspectral image.
Specifically, as shown in fig. 2, firstly, the number of end members is p, and the hyperspectral image is reduced to p-1 dimension in step S1, wherein the dimension reduction method is to perform noise whitening processing on the original hyperspectral image by using the minimum noise separation transformation (MNF) positive and negative transformation, and take the first p-1 wave bands for inverse transformation to obtain a noise whitened image as the hyperspectral image R.
The hyperspectral image target detection method based on the spatial projection information quantity and the end member volume double constraints is characterized by comprising the following steps of: in the step S1, the method for initializing the end member matrix and the abundance matrix is to randomly select p points in the hyperspectral image R as initial end members E0And let the initial abundance matrix C0=0。
In the method provided by this embodiment, the hyperspectral image end member projection space information quantity feature in step S2 is an information quantity of a hyperspectral image pixel projected onto an orthogonal complement space of an end member hyperplane.
In the hyperspectral image target detection method based on the spatial projection information content and the end member volume double constraints, the hyperspectral image end member projection spatial information content characteristic calculation method in the step S2 is as follows:
A) projecting all pixels of the hyperspectral image R into an end member hyperplane formed by stretching end members, and expressing a matrix M after projection as M ═ I-PE) R, wherein PEProjection operator, denoted P, for end-member matrixE=E(ETE)-1ET。
B) Information content J of orthogonal complement space of projection image R to end member hyperplaneIC(E) The sum of all non-negative eigenvalues for the projection matrix M, i.e.: j. the design is a squareIC(E)=tr(∑M)=∑αiWherein ∑MCov (M) is the covariance matrix of the projection matrix M,αias a covariance ∑MThe ith non-negative eigenvalue of (c).
In step S3 of the method provided in this embodiment, the objective function for fusing the projection information amount based on the spatial information statistical model and the minimum end-member volume constraint based on the convex geometric model is as follows:
wherein R, E and C are respectively a hyperspectral image, an end member matrix and an abundance matrix, and lambda and eta are respectively regularization parameters used for balancing a model error value, end member simplex volume constraint and end member projection space information content,is a minimum end member volume constraint function, E is an end member matrix of the hyperspectral image R, C is an abundance matrix, p is the number of end members,is a p-dimensional row vector.
The optimization iteration method in step S3 of the method provided in this embodiment is: and (3) finding a process of enabling f (E, C) to reach a minimum value through an iterative optimization method.
The method comprises the following specific steps:
B) The projection gradient method is adopted for optimization iteration of the objective function, and the iteration rule is as follows:
wherein alpha isiAnd betaiIs the iterative step length of an end member matrix and an abundance matrix, and the method in the chapter utilizes the famous Armijo technology to calculate alphaiAnd betai。
C) And setting the iteration calculation termination condition as a maximum iteration time condition, wherein the selected maximum iteration time is 150.
D) Taking into account the sum as a constraint, replacing the matrices R and E in the formulae in (A) and (B) in said 4. steps byAndselecting sigma as 20;
E) and when the iteration termination condition is met, ending the iteration, and obtaining a final end member matrix E and an abundance matrix C through iterative operation to obtain the ground object composition of each pixel.
In the method S4 provided in this embodiment, the land feature composition of each pixel is determined based on the end member matrix and the abundance matrix, and the extracted end member spectrum is matched with the target spectrum by using a spectrum Angle matching method (SAM), so as to detect the specified target.
Example 2
In order to implement the above method, the present invention further provides a hyperspectral image target detection system, as shown in fig. 3, including:
the initialization module is used for carrying out dimensionality reduction on the hyperspectral image and initializing an end member matrix and an abundance matrix;
the projection module is used for carrying out projection processing on the hyperspectral image pixels subjected to the dimension reduction processing by utilizing the initialized end member matrix and the abundance matrix to obtain hyperspectral image end member projection space information quantity characteristics;
the function solving module is used for iteratively solving a pre-constructed objective function based on the hyperspectral image end member projection space information quantity characteristic quantity to obtain an end member matrix and an abundance matrix;
the ground object determination module is used for determining the ground object composition of each pixel based on the end member matrix and the abundance matrix;
and constructing the objective function based on the projection information quantity of the spatial information statistical model and the minimum end member volume constraint based on the convex surface geometric model.
The initialization module is specifically configured to:
the dimension reduction method comprises the steps of utilizing minimum noise separation transformation (MNF) to carry out noise whitening processing on an original hyperspectral image by forward and reverse transformation, taking the first p-1 wave bands to carry out inverse transformation, and obtaining a noise whitening image which is used as a hyperspectral image R. And randomly selecting p points in the hyperspectral image R as initial end members E0And let the initial abundance matrix C0=0。
Projection module, in particular for
A) Projecting all pixels of the hyperspectral image R into an end member hyperplane formed by stretching end members, and expressing a matrix M after projection as M ═ I-PE) R, wherein PEProjection operator, denoted P, for end-member matrixE=E(ETE)-1ET。
B) Information content J of orthogonal complement space of projection image R to end member hyperplaneIC(E) The sum of all non-negative eigenvalues for the projection matrix M, i.e.: j. the design is a squareIC(E)=tr(∑M)=∑αiWherein ∑MCov (M) is the covariance matrix of the projection matrix M, αiAs a covariance ∑MThe ith non-negative eigenvalue of (c).
The embodiment provides an objective function of projection information quantity based on a spatial information statistical model and minimum end member volume constraint based on a convex geometric model, which is as follows:
wherein R, E and C are respectively a hyperspectral image, an end member matrix and an abundance matrix, and lambda and eta are respectively regularization parameters used for balancing a model error value, end member simplex volume constraint and end member projection space information content,is a minimum end member volume constraint function, E is an end member matrix of the hyperspectral image R, C is an abundance matrix, pThe number of the end members is the number of the end members,is a p-dimensional row vector.
The function solving module is specifically configured to:
B) The projection gradient method is adopted for optimization iteration of the objective function, and the iteration rule is as follows:
wherein alpha isiAnd betaiIs the iterative step length of an end member matrix and an abundance matrix, and the method in the chapter utilizes the famous Armijo technology to calculate alphaiAnd betai。
C) And setting the iteration calculation termination condition as a maximum iteration time condition, wherein the selected maximum iteration time is 150.
D) Taking into account the sum as a constraint, replacing the matrices R and E in the formulae in (A) and (B) in said 4. steps byAndselecting sigma as 20;
E) when the iteration termination condition is met, ending the iteration, obtaining a final end member matrix E and an abundance matrix C through iterative operation, and obtaining the ground object composition of each pixel
The ground object determination module is specifically configured to: and matching the extracted end member spectrum with the target spectrum by using a spectrum Angle matching method (SAM) to detect the specified target.
According to the hyperspectral image target detection method and system, a spectrum space information statistical model and a convex geometric model are fused to a non-negative matrix decomposition hyperspectral image end member generation method, so that the influence of interference information such as outliers and noise on end member extraction is reduced. The method is based on orthogonal subspace projection theory to carry out end member statistical information modeling, so that a space formed by a real end member which does not contain an external point has larger statistical information amount. The method takes the spatial statistical information and the end-member simplex volume as constraints and introduces the constraints into a non-negative matrix decomposition method, and constructs an objective function integrating model errors, the spatial statistical information quantity and the end-member simplex volume, so that the algorithm can perform iterative updating in a hyperplane formed by real end members. The method can effectively reduce the influence of interference factors such as outliers and noise, and improves the accuracy of end member extraction.
The present embodiment is only for explaining the present invention, and it is not limited to the present invention, and those skilled in the art can make modifications of the present embodiment without inventive contribution as needed after reading the present specification, but all of them are protected by patent law within the scope of the claims of the present invention.
Claims (10)
1. A hyperspectral image target detection method is characterized by comprising the following steps:
performing dimensionality reduction on the hyperspectral image, and initializing an end member matrix and an abundance matrix;
performing projection processing on the hyperspectral image pixels subjected to dimensionality reduction by using the initialized end member matrix and the abundance matrix to obtain hyperspectral image end member projection space information quantity characteristics;
iteratively solving a pre-constructed objective function based on the hyperspectral image end member projection space information quantity characteristic quantity to obtain an end member matrix and an abundance matrix;
determining the ground feature composition of each pixel based on the end member matrix and the abundance matrix;
and constructing the objective function based on the projection information quantity of the spatial information statistical model and the minimum end member volume constraint based on the convex surface geometric model.
2. The object detection method according to claim 1, characterized in that: the dimensionality reduction processing of the hyperspectral image comprises the following steps:
setting the number of end elements as p;
and performing noise whitening processing on the hyperspectral image by using the positive and negative transformation of the minimum noise separation transformation MNF, and performing inverse transformation on the first p-1 wave bands to obtain a noise whitened image serving as the hyperspectral image R after dimension reduction.
3. The object detection method according to claim 2, characterized in that: the initializing end member matrix and the abundance matrix comprise:
and randomly selecting p points in the hyperspectral image R as initial end members, and setting the initial value of the initial abundance matrix to be 0.
4. The object detection method according to claim 2, characterized in that: the method for obtaining the hyperspectral image end member projection space information quantity characteristics by projecting the hyperspectral image pixels subjected to dimensionality reduction by using the initialized end member matrix and the initialized abundance matrix comprises the following steps:
projecting all pixels in the hyperspectral image R to an end member matrix to obtain a projection matrix;
and obtaining the projection space information quantity characteristic of the end member of the hyperspectral image based on the sum of all the non-negative characteristic values in the projection matrix.
5. The object detection method according to claim 4, characterized in that: the projection matrix is calculated as follows:
M=(I-PE)R
where M is the projection matrix, PEA projection operator of the end member matrix E;
the projection operator PEIs calculated as follows:
PE=E(ETE)-1ET。
6. the object detection method according to claim 4, characterized in that: the calculation formula of the hyperspectral image end member projection space information quantity characteristic is as follows:
JIC(E)=tr(∑M)=∑αi
in the formula, JIC(E) Projecting spatial information quantity characteristics for the hyperspectral image end members, wherein M is a projection matrix; sigmaMIs the covariance matrix of the projection matrix M, alphaiAs a covariance matrix sigmaMThe ith non-negative eigenvalue of (c).
7. The object detection method according to claim 1, characterized in that: the objective function is:
E≥0,C≥0
in the formula (I), the compound is shown in the specification,for the model error values, λ and η are the regularization parameters, J, respectivelyVC(E) Is a minimum end member volume constraint function, E is an end member matrix of the hyperspectral image R, C is an abundance matrix, p is the number of end members,is a p-dimensional row vector;
the minimum end member volume constraint function is as follows:
8. the object detection method according to claim 7, characterized in that: the iterative solution of the pre-constructed objective function to obtain the end member matrix and the abundance matrix comprises the following steps:
Extracting an objective function optimization iteration rule based on the information content and the volume constraint non-negative matrix decomposition end member, and setting an iteration calculation termination condition;
and setting a sigma value, carrying out iterative solution on the objective function, and ending the iteration when an iteration termination condition is met.
Preferably: the objective function optimization iteration rule is as follows:
in the formula, Ei+1End-member matrix generated for the i +1 th iteration of the objective function, Ci+1Abundance matrix generated for the ith iteration, αiAnd betaiThe iteration step sizes of the end member matrix and the abundance matrix are respectively.
9. The object detection method according to claim 1, characterized in that: the determining of the terrestrial composition of each pixel based on the end member matrix and the abundance matrix comprises the following steps:
carrying out spectrum matching on the spectrum corresponding to the end member matrix and a preset target spectrum to obtain a target object corresponding to each end member matrix;
and determining the composition of the ground objects included in the pixel elements based on the abundance matrix and the target objects corresponding to the end member matrixes.
10. A hyperspectral image target detection system, comprising:
the initialization module is used for carrying out dimensionality reduction on the hyperspectral image and initializing an end member matrix and an abundance matrix;
the projection module is used for carrying out projection processing on the hyperspectral image pixels subjected to the dimension reduction processing by utilizing the initialized end member matrix and the abundance matrix to obtain hyperspectral image end member projection space information quantity characteristics;
the function solving module is used for iteratively solving a pre-constructed objective function based on the hyperspectral image end member projection space information quantity characteristic quantity to obtain an end member matrix and an abundance matrix;
the ground object determination module is used for determining the ground object composition of each pixel based on the end member matrix and the abundance matrix;
and constructing the objective function based on the projection information quantity of the spatial information statistical model and the minimum end member volume constraint based on the convex surface geometric model.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010692849.5A CN112036235A (en) | 2020-07-17 | 2020-07-17 | Hyperspectral image target detection method and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010692849.5A CN112036235A (en) | 2020-07-17 | 2020-07-17 | Hyperspectral image target detection method and system |
Publications (1)
Publication Number | Publication Date |
---|---|
CN112036235A true CN112036235A (en) | 2020-12-04 |
Family
ID=73579843
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010692849.5A Pending CN112036235A (en) | 2020-07-17 | 2020-07-17 | Hyperspectral image target detection method and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112036235A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112712028A (en) * | 2020-12-30 | 2021-04-27 | 闽江学院 | Spectrum unmixing method based on normalized ground object subspace projection |
CN116129281A (en) * | 2023-04-18 | 2023-05-16 | 中国人民解放军战略支援部队航天工程大学 | Sub-pixel target detection system for hyperspectral image |
CN116297391A (en) * | 2023-02-23 | 2023-06-23 | 北京市农林科学院 | Quick detection method for microplastic based on image recognition |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105320959A (en) * | 2015-09-30 | 2016-02-10 | 西安电子科技大学 | End member learning based hyperspectral image sparse unmixing method |
CN109241843A (en) * | 2018-08-02 | 2019-01-18 | 南京理工大学 | Sky spectrum joint multiconstraint optimization nonnegative matrix solution mixing method |
-
2020
- 2020-07-17 CN CN202010692849.5A patent/CN112036235A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105320959A (en) * | 2015-09-30 | 2016-02-10 | 西安电子科技大学 | End member learning based hyperspectral image sparse unmixing method |
CN109241843A (en) * | 2018-08-02 | 2019-01-18 | 南京理工大学 | Sky spectrum joint multiconstraint optimization nonnegative matrix solution mixing method |
Non-Patent Citations (5)
Title |
---|
COHEN, YUVAL等: "Sub-pixel target detection using local spatial information in hyperspectral images", 《CONFERENCE ON ELECTRO-OPTICAL REMOTE SENSING, PHOTONIC TECHNOLOGIES, AND APPLICATIONS V》 * |
DU, Q等: "Performance analysis for CEM and OSP", 《CONFERENCE ON ALGORITHMS AND TECHNOLOGIES FOR MULTISPECTRAL, HYPERSPECTRAL, AND ULTRASPECTRAL IMAGERY VIII》 * |
HARSANYI, JC等: "HYPERSPECTRAL IMAGE CLASSIFICATION AND DIMENSIONALITY REDUCTION - AN ORTHOGONAL SUBSPACE PROJECTION APPROACH", 《IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING》 * |
MIAO,LIDIAN;QI, HAIRONG: "Endmember extraction from highly mixed data using minimum volume constrained nonnegative matrix factorization", 《IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING》 * |
YUE SHI , HONGQI WANG等: "Endmember Extraction Using Minimum Volume and Information Constraint Nonnegative Matrix Factorization", 《IEEE GEOSCIENCE AND REMOTE SENSING LETTERS》 * |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112712028A (en) * | 2020-12-30 | 2021-04-27 | 闽江学院 | Spectrum unmixing method based on normalized ground object subspace projection |
CN112712028B (en) * | 2020-12-30 | 2024-04-09 | 闽江学院 | Spectrum unmixing method based on normalized ground object subspace projection |
CN116297391A (en) * | 2023-02-23 | 2023-06-23 | 北京市农林科学院 | Quick detection method for microplastic based on image recognition |
CN116297391B (en) * | 2023-02-23 | 2024-03-19 | 北京市农林科学院 | Quick detection method for microplastic based on image recognition |
CN116129281A (en) * | 2023-04-18 | 2023-05-16 | 中国人民解放军战略支援部队航天工程大学 | Sub-pixel target detection system for hyperspectral image |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Zhou et al. | A Gaussian mixture model representation of endmember variability in hyperspectral unmixing | |
CN109522956B (en) | Low-rank discriminant feature subspace learning method | |
CN108537102B (en) | High-resolution SAR image classification method based on sparse features and conditional random field | |
CN112036235A (en) | Hyperspectral image target detection method and system | |
Halimi et al. | Fast hyperspectral unmixing in presence of nonlinearity or mismodeling effects | |
EP2291794B1 (en) | Image processing | |
Fawzi et al. | Measuring the effect of nuisance variables on classifiers | |
CN109190511B (en) | Hyperspectral classification method based on local and structural constraint low-rank representation | |
Tao et al. | A new deep convolutional network for effective hyperspectral unmixing | |
CN111368691B (en) | Unsupervised hyperspectral remote sensing image space spectrum feature extraction method | |
Rezaei et al. | Endmember extraction using a combination of orthogonal projection and genetic algorithm | |
Cheng et al. | ANSGA-III: A multiobjective endmember extraction algorithm for hyperspectral images | |
CN109063555B (en) | Multi-pose face recognition method based on low-rank decomposition and sparse representation residual error comparison | |
CN115410088A (en) | Hyperspectral image field self-adaption method based on virtual classifier | |
Uezato et al. | Incorporating spatial information and endmember variability into unmixing analyses to improve abundance estimates | |
Ge et al. | Adaptive hash attention and lower triangular network for hyperspectral image classification | |
Deng et al. | An infinite Gaussian mixture model with its application in hyperspectral unmixing | |
CN111008652A (en) | Hyper-spectral remote sensing image classification method based on GAN | |
Guo et al. | A modified huber nonnegative matrix factorization algorithm for hyperspectral unmixing | |
Shahid et al. | Spatial-aware hyperspectral nonlinear unmixing autoencoder with endmember number estimation | |
CN113139515A (en) | Hyperspectral image classification method based on conditional random field and depth feature learning | |
CN111460943A (en) | Remote sensing image ground object classification method and system | |
Kong et al. | Regularized multiple sparse Bayesian learning for hyperspectral target detection | |
Miao et al. | Sparse unmixing for hyperspectral imagery via comprehensive-learning-based particle swarm optimization | |
CN115496950A (en) | Neighborhood information embedded semi-supervised discrimination dictionary pair learning image classification 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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20201204 |