CN-117953246-B - Hyperspectral mixed pixel decomposition method and device based on end member spectrum variability

CN117953246BCN 117953246 BCN117953246 BCN 117953246BCN-117953246-B

Abstract

The invention relates to the technical field of hyperspectral remote sensing data processing, in particular to a hyperspectral mixed pixel decomposition method and device based on end member spectrum variability. The method comprises the steps of constructing a hyperspectral mixed pixel decomposition error model based on end member spectrum variability based on a traditional unmixing model, determining the statistical distribution of abundance inversion errors based on end member spectrum statistical distribution and hyperspectral mixed pixel decomposition error model used for representing the end member spectrum variability, deriving the statistical distribution of real abundance based on the statistical distribution of abundance inversion errors, and carrying out end member spectrum extraction on a hyperspectral image to be unmixed so as to determine the confidence interval of the ground object abundance of the hyperspectral image based on the statistical distribution of real abundance.

Inventors

XU HUIHUA
XIU PENG
CHEN WEILI

Assignees

北京环境特性研究所

Dates

Publication Date: 20260512
Application Date: 20240119

Claims (10)

1. The hyperspectral mixed pixel decomposition method based on end member spectrum variability is characterized by comprising the following steps of: based on a traditional unmixing model, constructing a hyperspectral mixed pixel decomposition error model based on end member spectrum variability; Determining a statistical distribution of abundance inversion errors based on an end member spectrum statistical distribution used for representing end member spectrum variability and the hyperspectral mixed pixel decomposition error model; deriving a statistical distribution of true abundance based on the statistical distribution of abundance inversion errors; and carrying out end member spectrum extraction on the hyperspectral image to be unmixed so as to determine the confidence interval of the ground object abundance of the hyperspectral image based on the statistical distribution of the real abundance.
2. The method of claim 1, wherein constructing a hyperspectral mixed pel decomposition error model based on end-member spectral variability based on the traditional unmixing model comprises: Based on a traditional unmixing model, constructing a hyperspectral mixed pixel decomposition model taking the variability of the end member spectrum into consideration; And performing difference between the hyperspectral mixed pixel decomposition model and the traditional unmixing model, and adopting a least square method to perform fitting solution to obtain a hyperspectral mixed pixel decomposition error model for representing the quantitative relation between the abundance inversion error and the end member spectrum estimation error.
3. The method of claim 2, wherein the hyperspectral hybrid pel decomposition model is: wherein R epsilon R L×1 is the spectrum of the hyperspectral mixed pixel, The end-member spectrum matrix used for the unmixing of the traditional unmixing model, For the abundance fraction of various ground objects unmixed by using the traditional unmixed model, e epsilon R L×1 is an error term, R is a real number of 0-1, L is a wave band number, p is a ground object type number, M is a real spectrum of various ground objects, f is a real abundance of various ground objects, The error is estimated for the end-member spectrum, The error is inverted for abundance.
4. The method of claim 2, wherein the hyperspectral hybrid pel decomposition error model is: Wherein Deltaf epsilon R p×1 is an abundance inversion error, G SCLS is a coefficient matrix obtained by solving a least square method, deltaM epsilon R L×p is an end member spectrum estimation error, In order to use the abundance fraction of various ground objects obtained by the traditional unmixing model, I is a unit array, The end-member spectrum matrix used for the unmixing of the traditional unmixing model, Is that C= [1, ], 1] T ∈R p×1 is a vector of all 1, R is a real number of 0-1, L is a band number, and p is a ground object type number.
5. The method of claim 1, wherein the statistical distribution of abundance inversion errors is: Wherein Deltaf epsilon R p×1 is an abundance inversion error, N () represents normal distribution, 0 is zero vector, sigma Δf is a covariance matrix of the abundance inversion error, A, B is an intermediate variable, nonsensical, G SCLS is a coefficient matrix obtained by solving a least square method, Abundance fractions of various features unmixed using a conventional unmixed model Sigma ii is the autocovariance matrix of the i-th class of end-member spectra, sigma ij is the cross-covariance matrix between the i-th and j-th classes of end-member spectra, and p is the number of ground object classes, i.e., the number of end-member spectrum classes.
6. The method of claim 1, wherein the statistical distribution of true abundances is: wherein f is the real abundance of various ground features, N () represents normal distribution, In order to use the abundance fraction of various ground objects which are unmixed by a traditional unmixed model, sigma Δf is a covariance matrix of abundance inversion errors, A, B is an intermediate variable, nonsensical, G SCLS is a coefficient matrix obtained by solving a least square method, Abundance fractions of various features unmixed using a conventional unmixed model Sigma ii is the autocovariance matrix of the I-th class of end-member spectrum, sigma ij is the cross-covariance matrix between the I-th class and the j-th class of end-member spectrum, p is the number of ground object types, i.e. the number of end-member spectrum types, I is a unit matrix, The end-member spectrum matrix used for the unmixing of the traditional unmixing model, Is that C= [1, ], 1] T ∈R p×1 is a vector of all 1.
7. The method of claim 1, wherein the performing end member spectral extraction on the hyperspectral image to be unmixed to determine a confidence interval for the feature abundance of the hyperspectral image based on the statistical distribution of true abundances comprises: Performing end member spectrum extraction on a hyperspectral image to be unmixed by using an automatic end member extraction algorithm to obtain an end member spectrum set of the hyperspectral image, wherein each type of end member spectrum in the end member spectrum set corresponds to one type of ground object; based on the end member spectrum set, estimating an overall mean value, an autocovariance matrix and a cross covariance matrix of each type of end member spectrum of the hyperspectral image to obtain a corresponding estimation result; determining an end member spectrum matrix used by unmixing of a traditional unmixing model based on an estimation result of the overall mean value of each type of end member spectrum; Solving abundance fractions of various ground objects unmixed by using the traditional unmixed model and an end member spectrum matrix used by the unmixed of the traditional unmixed model; Carrying the estimation results of the auto-covariance matrix and the cross-covariance matrix of each type of end member spectrum, the end member spectrum matrix used by the unmixing of the traditional unmixing model and the abundance fraction of various types of ground objects unmixed by the traditional unmixing model into the statistical distribution of the real abundance to obtain the statistical distribution of the real abundance of the hyperspectral image; and obtaining a confidence interval of the ground object abundance of the hyperspectral image based on the statistical distribution of the real abundance of the hyperspectral image.
8. A hyperspectral mixed pel decomposition device based on end member spectral variability, comprising: the construction unit is used for constructing a hyperspectral mixed pixel decomposition error model based on end member spectrum variability based on the traditional unmixing model; the determining unit is used for determining the statistical distribution of the abundance inversion error based on the end member spectrum statistical distribution used for representing the variability of the end member spectrum and the hyperspectral mixed pixel decomposition error model; a deriving unit for deriving a statistical distribution of true abundances based on the statistical distribution of abundance inversion errors; The decomposition unit is used for carrying out end member spectrum extraction on the hyperspectral image to be unmixed so as to determine the confidence interval of the ground object abundance of the hyperspectral image based on the statistical distribution of the real abundance.
9. A computing device comprising a memory and a processor, the memory having stored therein a computer program, the processor implementing the method of any of claims 1-7 when the computer program is executed.
10. A computer readable storage medium having stored thereon a computer program which, when executed in a computer, causes the computer to perform the method of any of claims 1-7.

Description

Hyperspectral mixed pixel decomposition method and device based on end member spectrum variability Technical Field The embodiment of the invention relates to the technical field of hyperspectral remote sensing data processing, in particular to a hyperspectral mixed pixel decomposition method and device based on end member spectrum variability. Background With the development of remote sensing science and technology, hyperspectral remote sensing has become an important technology for earth observation, can acquire a characteristic curve with high spectral resolution reflecting the distribution attribute of ground objects pixel by pixel while imaging the ground surface, is widely applied to ground object detection and ground surface parameter inversion, and has been successfully applied to aspects of geological mapping, accurate agriculture, ecological environment monitoring and the like. However, due to complexity of ground object distribution and limitation of spatial resolution of a hyperspectral imager, mixed pixels are ubiquitous in hyperspectral remote sensing data. Hyperspectral mixed pixel decomposition or spectral unmixing aims at decomposing mixed pixels into end member spectra characterizing different classes of materials and their corresponding abundance fractions. However, in most application scenarios, the premise of end member spectrum fixation is assumed to be unsatisfied and theoretically fixed spectra do not exist. On the one hand, the difference of the spectrum amplitude or shape of the end member spectrum in time and space is caused by the differences of atmospheric conditions, terrains and illumination conditions, surrounding environment and other factors, and on the other hand, the classification of the ground object types is usually macroscopic and problem-oriented, and is related to the classification fineness of the ground object, so that the similar ground object possibly contains a plurality of subclasses, and the inherent physicochemical property differences exist, and the end member spectrum also has a certain amplitude. Both of these causes ultimately lead to the presence of "isospectral" and "foreign isospectral" phenomena in the end-member spectra, referred to as end-member variability. The traditional hyperspectral mixed pixel decomposition method generally adopts the typical pure pixel spectrum of various ground objects or the average spectrum of a plurality of pure pixel spectrums as a fixed end member spectrum, and the mixed pixels in a scene are decomposed one by one, and the abundance fraction of various ground objects in each pixel is obtained through inversion. It can be seen that the conventional hyperspectral mixed pel decomposition method does not consider the end member spectrum variability, so that the conventional hyperspectral mixed pel decomposition method is poor in accuracy. Therefore, there is a need for a hyperspectral mixed pel decomposition method based on end-member spectral variability. Disclosure of Invention In order to solve the problem of poor accuracy of a traditional hyperspectral mixed pixel decomposition method, the embodiment of the invention provides a hyperspectral mixed pixel decomposition method and device based on end member spectrum variability. In a first aspect, an embodiment of the present invention provides a method for decomposing hyperspectral mixed pixels based on end-member spectrum variability, including: based on a traditional unmixing model, constructing a hyperspectral mixed pixel decomposition error model based on end member spectrum variability; Determining a statistical distribution of abundance inversion errors based on an end member spectrum statistical distribution used for representing end member spectrum variability and the hyperspectral mixed pixel decomposition error model; deriving a statistical distribution of true abundance based on the statistical distribution of abundance inversion errors; and carrying out end member spectrum extraction on the hyperspectral image to be unmixed so as to determine the confidence interval of the ground object abundance of the hyperspectral image based on the statistical distribution of the real abundance. In a second aspect, an embodiment of the present invention further provides a hyperspectral mixed pel decomposition device based on end-member spectrum variability, including: the construction unit is used for constructing a hyperspectral mixed pixel decomposition error model based on end member spectrum variability based on the traditional unmixing model; the determining unit is used for determining the statistical distribution of the abundance inversion error based on the end member spectrum statistical distribution used for representing the variability of the end member spectrum and the hyperspectral mixed pixel decomposition error model; a deriving unit for deriving a statistical distribution of true abundances based on the statistical distribution of abundance inversion errors