Linear Spectral Unmixing function

Available with Image Analyst license.

Available with Spatial Analyst license.

Overview

Performs subpixel classification and calculates the fractional abundance of different land-cover types for individual pixels.

Notes

The Linear Spectral Unmixing function calculates the fractional cover for individual pixels that contain multiple land cover types. It generates a multiband layer, where each band corresponds to the fractional abundance of each land cover class. For example, you can use it to perform land cover classification on a multispectral image to identify photosynthetic vegetation, bare soil, and dead or nonphotosynthetic vegetation.

The order of the output multiband raster follows the order of the input spectral profile.

The number of classes cannot exceed the number of bands in the input raster. For example, you cannot extract information about 9 classes from an 8-band raster.

Polygon features require the following field names:

• classname—A text field indicating the name of the class category.
• classvalue—A long integer field containing the integer value for each class category.

When calculating the fractional abundance of each land cover class, the solution to the system of equations can include negative coefficients or fractions. If this occurs, you can check the training samples in your input spectral profile to see that they accurately represent each class. If they appear correct, select Non-negative for Output Value Option.

This function can process a multidimensional raster layer. The result is a multidimensional raster layer where each slice is a multiband raster containing the fractional abundance of each land cover class.

When the Remove Continuum check box is checked, a convex hull is fitted over the top of a spectrum using straight-line segments that connect local spectra maxima. The continuum will be removed by dividing it into the actual spectrum for each pixel in the image and the input spectra. This allows comparison of absorption features from a common baseline.

Parameters

Parameter name	Description
Raster	The input raster.
Spectral Profile Type	The spectral information for the different land cover classes. Spectral Profile—A classifier definition file (.ecd), generated from the Train Maximum Likelihood Classifier tool, or a JavaScript Object Notation file (.json) that contains the class spectral profiles. Training Feature—Polygon features or a training sample feature class, generated from the Training Samples Manager.
Spectral Profile	The input spectral profile file.
Training Feature	The input training feature file. Supported file formats include .esl, .sli, .ecd, and .json.
Non-negative	Specifies the options to define the output pixel values. Checked—There will be no negative output values. Unchecked—There can be negative values of fractional land cover.
Sum to One	Specifies the options to define the output pixel values. Checked—Class values for each pixel are provided in decimal format with the sum of all classes equal to 1—for example, Class1 = 0.16; Class2 = 0.24; Class3 = 0.60. Unchecked—The sum of all classes in a pixel can exceed 1.
Remove Continuum	The spectra will be normalized from an image or reference data. Checked—The continuum will be removed to normalize the spectra. Unchecked—The continuum will not be removed.