Data Exploration and Analysis of Hyperspectral images: Visualization and Symbolic Description
In order to give a definition of features which are helpful for dimensionality reduction, it is necessary to explore the data set via the visualization or symbolic representation. The major objectives of data analysis are to summarize and interpret a data set, describing the contents and exposing important features. Visualization can play an important role in data exploring and analysis. For example, the image cube is often used to represent the whole data set of the hyperspectral images. Several graphical methods have been proposed for visualizing high-dimensional data items directly, by mapping the apparent data values of each dimension into one figure. Data projection is the common visual way to get the interesting subsets of the original data, and certain properties of the structure of the data sets can be preserved as faithfully as possible. In this paper, some projection methods are used to explore and analyze the hyperspectral images. In addition, the multi-scale approaches are used to visualize the
hyperspectral curve in a time-scale plane. The useful features then can be explored and extracted for further applications. An AVIRIS data set with five classes is used to demonstrate the ways of visualization and symbolic description.
2. Visualization Techniques
For the convenience of analyzing and quantifying the characteristics of hyperspectral data, it is necessary to define mathematically and conceptually some representation spaces to inspect the data variations from some aspects. Landgrebe (1997) illustrated that there have been three principle ways in which multispectral data are represented quantitatively and visualized. Figure 1 shows three different data spaces which are used to represent multispectral images. The same representations are still convenient for hyperspectral data. In this paper, some extended methods of the three visualization techniques are described to characterize the hyperspectral images.
Figure 1. Three data space for representing multispectral data
2.1 Spatial-Spectral Space
Data in the RGB image space (see figure 1.a) directly offer a visual way to understand the spatial variation of the scene and the relationship between an individual pixel and the land cover class it belongs to. Tasks of manual image interpretation are usually carried out in the image space. However, the RGB image only shows the spatial information of three bands corresponding to Red, Green and Blue. In contrast to RGB images, the spectral slices are used to extract a combined spatial/spectral profile from a hyperspectral image. Figure 2.a shows a spectral slice with two classes. The vertical direction corresponds to the spatial dimension of the image being slices, the horizontal direction corresponds to the spectral dimension, and the grayscale shows the spectral density (reflectance, radiance, etc.). By a perspective view, a which creates a RGB image with the spectral slice of the top row and right column can be displayed on the 2D screen. Figure 2.b shows a hyperspectral image cube.
Figure 2. Spatial-spectral space
