The classifier implements methods to detect patterns in image regions. Commonly, classification algorithms are divided by the level of classification (pixel or region), and by the interaction of the user (supervised or unsupervised). Pixel-based algorithms classify individual pixels according to their resemblance to a specific pattern. Region based algorithms use regions from segmented images and classify each region to a specific pattern. The supervised method uses a predefined typology, given by the user, who supplies samples of each pattern. Unsupervised methods detect an unknown number of patterns, according to their own method.
The available methods are:
ISOSeg
This is an unsupervised and region-based classification algorithm to classify regions in a segmented image applied on a region set that will be characterized by its statistical attributes of media, covariance matrix and area.
Input:
Raster
Vector of polygons
Acceptance threshold
K-Means
This is an unsupervised and pixel-based classification algorithm.
Input:
Raster
The value of “K”, which stands for the number of patterns to find in the image.
A convergence threshold. When the clusters move less than this threshold, the algorithm stops.
Maximum number of iterations.
Expectation-Maximization - EM
This is an unsupervised and pixel-based classification algorithm. Expectation-Maximization (EM) works interactively by applying two steps: the E-step (Expectation) and the M-step (Maximization).
Input:
Raster
The value of “K”, which stands for the number of patterns (or clusters) to find in the image.
The maximum number of iterations (E/M steps) to perform if convergence is not achieved.
The maximum number of points used to estimate the clusters (default = 1000).
A convergence threshold. When the clusters change in a value smaller then epsilon, the convergence is achieved.
The previously estimated means of the clusters (optional).
Spectral Angle Mapper - SAM
This is an supervised classification algorithm.
Input:
Raster
A set of ROI samples.
Maximum a Posteriori Probability - MAP
This is an supervised classification algorithm.
Input:
Raster
A set of ROI samples.
Select PROCESSING → RASTER PROCESSING → CLASSIFIER in the main menu and adjust the parameters.
On the List of Layers select the raster layer to apply the operation.
Optionally use Filter By Name field giving part of the layer name to help find the layer in the list.
Press the Next button to go to next step or the Cancel button to close the dialog.
Classification parameters
Select the type of classifier to be used.
Select the bands to be used in the process.
As described above, each classifier has a set of specific attributes. For supervised classifiers (SAM and MAP) is necessary to use a component for the acquisition of samples.
Use the tool
to enable the acquisition of components (samples) through the
interface ROI Manager. This interface offers several tools (such as
create by polygon
and create by extent
)
to help on collecting good samples.
The user can load a
saved ROI set. Click on the icon
and browse the *.shp file.
Label: enter the
name of the class and select the color to it. Click on the icon
to include it in the ROI set.
Click on the created class and select one of the right tools to create the samples.
A created ROI set
can be saved using the Export ROI Set. Click on the icon
and enter a name to the *.shp file. To export it click on the icon
.
Click on the Close button.
Press the Next button to go to next step, the Back button to return to the previous wizard or the Cancel button to close the dialog.
Output information
Raster Info -
First press
and inform the folder where the resulting file will be saved.
Name - inform the raster name.
Extra Parameters - if there are some.
Press the Finish button to save the resulting classified raster or the Back button to go to the previous page.