Self-adaptive hyperspectral image unmixing method based on region segmentation

A hyperspectral image and region segmentation technology, applied in the field of image processing, can solve problems such as inapplicable hyperspectral data

CN104952050AActive Publication Date: 2015-09-30XIDIAN UNIV
2 Cites 31 Cited by

Patent Information

Authority / Receiving Office
CN · China
Current Assignee / Owner
Publication Date
2015-09-30

Smart Images

  • Figure 1
    Figure 1
  • Figure 2
    Figure 2
  • Figure 3
    Figure 3
Patent Text Reader

Abstract

The invention discloses a self-adaptive hyperspectral image unmixing method based on region segmentation. In consideration of coexistence of linear mixing and bilinear mixing, the method is implemented by adopting the following steps: inputting a hyperspectral image; estimating the number of end elements with a minimum error based hyperspectral signal recognition method; extracting end element matrixes with a vertex component analysis algorithm; clustering hyperspectral data with a K-means clustering method, and segmenting the image into a homogeneous region and a detail region; adopting a linear model for the homogeneous region and performing unmixing with a sparse-constrained non-negative matrix factorization method, and adopting a generalized bilinear model for the detail region and performing unmixing with a sparse-constrained semi-non-negative matrix factorization method. According to the method, characteristics of the hyperspectral data and abundance are combined, the hyperspectral image is represented more accurately, and the unmixing accuracy rate is increased. The sparse constraint condition is added to the abundance, the defect of high probability of local minimum limitation of the semi-non-negative matrix factorization method is overcome, more accurate abundance is obtained, and the method is applied to ground-object recognition for the hyperspectral image.
Need to check novelty before this filing date? Find Prior Art

Description

technical field

[0001] The invention belongs to the technical field of image processing, and mainly relates to an unsupervised target recognition method, in particular to a hyperspectral image self-adaptive unmixing method based on region segmentation, which can be applied to ground object recognition of hyperspectral images. Background technique

[0002] Hyperspectral remote sensing uses the principle of spectroscopy, that is, in the ultraviolet, visible, near-infrared and mid-infrared regions of the electromagnetic spectrum, many very narrow and spectrally continuous image data are obtained by imaging spectrometers. Imaging spectrometers acquire ground reflection or emission spectral signals in units of pixels. The object spatial region corresponding to each pixel in the image often contains different substances with different spectral characteristics. If the pixel contains only one substance or the proportion of the substance is very high, it is called a pure pixel, also...

Examples

Embodiment 1

[0048] The present invention is a hyperspectral image adaptive unmixing method based on region segmentation, referring to figure 1 , the specific implementation steps of the present invention include:

[0049] (1) Input hyperspectral image, hyperspectral image data Y∈R L×N where L represents the number of bands of hyperspectral data, N represents the total number of hyperspectral data samples, and R represents the real number domain; in this example, the rgb image of the 30th band of the input hyperspectral image is as follows figure 2 As shown, the figure contains three substances: vegetation, water and soil. Hyperspectral unmixing is to obtain the percentages of these three different substances, that is, the abundance map.

[0050] (2) Estimate hyperspectral data Y∈R using the minimum error hyperspectral signal recognition method L×N The signal subspace of the signal subspace, the dimension K of the signal subspace is obtained, that is, the number of endmembers of the hy...

Embodiment 2

[0060] The hyperspectral image adaptive unmixing method based on region segmentation is the same as embodiment 1, wherein step (5) described with L 1 / 2 Constrained non-negative matrix factorization method to obtain homogeneous regional data Y 1 The first-order abundance matrix X 1 , including the following steps:

[0061](5a) According to the hyperspectral imaging theory, the abundance matrix X of the data in the homogeneous region of the hyperspectral image 1 Add L to 1 / 2 Norm, get the sparse constraint expression as the abundance matrix X 1 The sparse constraint term, where x 1n (k) is the hyperspectral image homogeneous region data Y 1 The abundance of the kth endmember of the nth pixel in .

[0062] (5b) Add the sparse constraint item obtained in step (5a) to the objective function of the non-negative matrix factorization algorithm based on Euclidean distance , forming a new objective function:

[0063] m i n 1 ...

Embodiment 3

[0072] The hyperspectral image adaptive unmixing method based on region segmentation is the same as embodiment 1-2, wherein step (6) uses L 1 / 2 -Semi_NMF method to get detail area data Y 2 The corresponding first-order abundance matrix X 2 and the second-order abundance matrix E, according to the following steps:

[0073] (6a) In hyperspectral image detail area Y 2 The bilinear model is used to express as follows

[0074] Y 2 =AX 2 +BE+M

[0075] in, Represents the first-order abundance matrix corresponding to the data in the bilinear region, where each column vector represents the abundance vector of the nth pixel, is a bilinear endmember matrix, Is the second-order abundance matrix corresponding to the data in the bilinear region, where each column vector represents the bilinear abundance vector of the nth pixel, Represents the noise matrix;

[0076] (6b) The abundance matrix X of the data in the homogeneous region of the hyperspectral image 2 Add L to 1 / ...