SPECLASS

Description

SPECLASS creates a spectral classification from optical data by converting it to TOA reflectance and extracting per-pixel features and indices. It is a spectral pre-classification containing relevant metadata for radiometric normalization.

Parameters

Name	Type	Length	Value range
Input Layers: Input image *	Raster port	1 -
DEM Layer: Elevation channel or layer	Raster port	0 - 1

* Required parameter

Parameter descriptions

Input Layers: Input image

The name of the input image from a supported sensor.

Input file must contain at least red, green and NIR spectral channels. It has to contain enough metadata information to convert data to TOA reflectance.

DEM Layer: Elevation channel or layer

Specifies the elevation layer. The default is to take the first layer containing elevation data

Details

The algorithm is based on rules that use spectral indices and features derived from spectral bands. These features are calculated for a multitude of sensors and normalized so that a cross sensor comparison of features becomes possible.

Spectral bands are converted to TOA reflectance values on-the-fly, using the nominal solar zenith angle. Those values are used to calculate indices and features like tasseled cap indices, NDVI, HOT, MNDWI, and others.

Various levels of classification are processed based on the available number of bands. At a minimum, red, green and NIR bands must be available.

At the first classification level, only basic features can be derived, and a limited number of classes can be extracted, such as clouds, water and shadow, vegetation and bare ground.

At the second level of classification, a SWIR band is required, and basic classes are subdivided into subclasses, such as thick or thin clouds; snow; turbid, green, shallow or deep water; various types of active and senescent vegetation, and so forth.

The last level uses a cirrus band, if available, to extract a high-cloud class.

SPECLASS can be used for the generation of a general quality layer to guide further radiometric and geometric processing. The resulting cloud and water masks can be used for targeted ground control point and tie point collection leading to better image triangulation. In the same way it may improve the extraction and automatic filtering of extracted surface models filling water bodies with a constant elevation, ignoring clouds and filtering heavily in shadow. Using the cloud and water mask for the avoidance of dodging points in color balancing will lead to better mosaics.

The output from SPECLASS can be recoded to fulfill the requirements of a class layer for topographic normalization such as TOPOSOLNORM.

The possible output classes are as follows: