Automatic Denoising And Unmixing In Hyper-Spectral Image Processing

Read Complete Research Material

AUTOMATIC DENOISING AND UNMIXING IN HYPER-SPECTRAL IMAGE PROCESSING Automatic denoising and unmixing in Hyper-spectral image processing



Automatic denoising and unmixing in Hyper-spectral image processing

 

ABSTRACT

One of the problems of processing hyperspectral imaging is the identification of significant data. This problem is solved by means of classification. In order to solve classification problems, it is proposed the construction of maps, in which the spectra of different surfaces areas associated with different materials, is clearly divided. In this approach, we rely on the method of "diffusion map". The basic idea is that the multidimensional data are projected into the mathematical variety of small dimension while preserving the mutual relations between the data. We describe a diffuse process that reveals hidden patterns that exist between the spectra of separating the different layers of the background. The model is based on a random walk on a graph. Random walk divides the region into separate clusters, which are caused by hidden relationships between elements of the set. As a result of applied techniques of random processes in Markov chains for the separation of clusters.

Keywords: diffuse map, clustering; hyperspectral image Hyperspectral image - a cube of data that includes spatial information (2D) of the object, supplemented with the spectral information (1D) for each spatial coordinate. In other words, each point in the image corresponds to the spectrum obtained at this point of the subject.

One of the main tasks of processing hyperspectral image recognition is uncertain terrain areas as belonging to one or another known class. Recognition technology is represented by a set of basic algorithms: AdaBoost, SVM, Neural Networks, Linear Discriminate Analysis. After analyzing these algorithms, we can conclude that existing detection methods uncertain areas of hyperspectral images are not as effective. Therefore, the solution of problems of recognition of indeterminate zones hyperspectral images remain relevant.

In our case, the task of recognition of indeterminate zones of hyperspectral images is proposed to construct a mapping, in which the spectra of different surface areas associated with different materials, clearly split.

In this approach, we rely on the method of "diffuse maps" [3].

This method was first used for modeling three-dimensional objects on the basis of a plurality of two-dimensional projections of the object representations (pictures). The method consists in the fact that multidimensional data is projected into a low-dimensional mathematical manifold preserving the mutual relations between the data. In this topology diversity simulates the difference between projections. That is, the variation of the data is described by diversity, to build a diffuse map. In the case where the manifold is three-dimensional, it is a three-dimensional model projections.

In this article we describe a diffuse process that reveals hidden patterns that exist between the spectra of separating different layers background.

The model that we propose is based on the random walk on a graph.

Denote

The set of objects represented by multidimensional vectors. The first index denotes the numbering of objects in the classroom, the second index labels the classes ...