Purification and optimization-based space target spectrum de-mixing method

Abstract

The invention discloses a purification and optimization-based space target spectrum de-mixing method. The method comprises the following steps of: firstly extracting an end member of an original sparse spectrum by adoption of variable division augmented Lagrange algorithm through a purification method, wherein the purification method is capable of effectively selecting an initial end member of a mixed spectrum so as to decrease a calculated amount; carrying out iterative purification and optimization on the initially extracted end member by utilizing an end member purification and optimizationmethod, wherein the optimized end member approaches a real end member step by step; and finally comparing the purified and optimized end member and an abundance with the original spectrum, and continuously carrying out purification until a reconstruction error of the purified end member is greater than a minimum value of a previous error, wherein the obtain end member and the abundance are optimal solutions. The method has the advantages of being strong in robustness and high in speed, and is capable of carrying out correct solution under the condition of low information amount.

Description

technical field [0001] The invention relates to the field of space target recognition, in particular to a method for unmixing the spectrum of space targets based on purification and optimization. Background technique [0002] With the rapid development of modern science and technology, more and more platforms are sent into space, and the research on space targets (such as satellites, space debris, cosmic stars, etc.), including space target identification, orbit determination and position estimation, etc. It is becoming more and more important. It is the basis for entering space, understanding space, and controlling space. It is an indispensable link in space attack and defense. Obtaining the spectral information and characteristics of space objects is of great significance for the identification of space objects, orbit estimation, judgment of satellite attitude and working status, etc. [0003] As a new detection and analysis method, spectral analysis technology has been w...

AI Technical Summary

Problems solved by technology

Moreover, due to the limited time of the ground observation window, the obtained observation data is less, and the spectrum is affected by the atmosphere, radiation, temperature, etc., resulting in a low signal-to-noise ratio

In recent years, with the development of data processing and modeling, spectral object unmixing has been greatly developed, and spectral unmixing methods based on spectral linear mixture models have emerged, which are mainly divided into three categories: the first category is the pixel purity index Algorithm (Purepixel index, referred to as PPI), this algorithm uses geometric convex sets as the basis, reduces the dimensionality of the original data, randomly generates test vectors that pass through the data set as one-dimensional coordinate axes, and records that each pixel is projected to The number of endpoints of the one-dimensional test vector coordinates is recorded as the PPI index. The disadvantage is that the selection of the initial test vector has a great influence on the result

The second type is the N-norm algorithm (N‐FINDER), which determines the position of the endmember by calculating the volume of the simplex. This method randomly selects pixels from the data set as the initial candidate endmember, and calculates the Simplex volume, and then replace the initial candidate endmember with the observed pixel in the image, recalculate the simplex volume, repeat the process of substituting candidates, comparing, and replacing until the vertex of the largest simplex is found. The disadvantage is that the initial endmember The randomness of the endmember extraction results is uncertain and non-repeatable; when calculating the volume, the data needs to be reduced in dimensionality, and the dimensionality reduction operation will lead to the loss of weak information in the data; the N-FINDER algorithm has a long iteration time and is not efficient

The third type is vertex component analysis (VCA), which is based on the geometric description of the linear spectral mixture model, and finds the pixel with the largest projection length after projection as the first end member, and then uses the A direction orthogonal to the endmembers is used as the projection direction in the second iteration. After calculating P orthogonal projections, all endmembers can be obtained, and the endmember extraction is realized with a very low time complexity. The disadvantage is that the spectral data to be measured There must be an endmember in

[0005] To sum up, the existing methods have large requirements on the amount of spectral data, and it is difficult to apply to the unmixing of space objects with small amount of spectral data and sparse spectra; moreover, due to the limited window time for ground observation of space objects and the spectral Affected by the atmosphere, radiation, temperature, etc., the signal-to-noise ratio is low, and it is difficult for existing methods to effectively perform spectral unmixing in the case of low signal-to-noise ratio; greater impact

Images