RP → Mixture Model

This module implements raster decomposition using the mixture model strategy. The spatial resolution of remote sensing satellites, in general, allows that a single pixel contains more than a target (for example vegetation + shadow + soil). The mixture model algorithms allow to decompose the raster into fraction images, where the value of the resultant pixels indicate the fraction of each target inside the pixel.

Input:

• One Raster
• A map of pure endmembers, which are pixel values where the user knows the exact proportion of a component.
• Sensor information for each band, if available

The available methods in TerraLib are:

• Linear
• Principal Component Analysis (PCA)

It is calculated by the following equation: $$r_i = \sum_{j = 1}^n a_{ij} x_j + e_i$$ where

• $r_i$ is the pixel value at band $i$
• $a_{ij}$ is the known value of the component $j$ at band $i$
• $x_j$ is the proportion of component $j$
• $e_i$ is the estimation error for band $i$

The component matrix is transformed using Principal Component Analysis, and the fraction of each component is obtained by using the proper number of eigenvectors.

// open input raster
std::map<std::string, std::string> rinfo;
rinfo["URI"] = "input.tif";
te::rst::Raster* rin = te::rst::RasterFactory::open(rinfo);
 
// create algorithm parameters
te::rp::MixtureModel::InputParameters mmInputParameters;
te::rp::MixtureModel::OutputParameters mmOutputParameters;
 
// defining input parameters
mmInputParameters.m_inputRasterPtr = rin;
for (unsigned b = 0; b < rin->getNumberOfBands(); b++)
  mmInputParameters.m_inputRasterBands.push_back(b);
mmInputParameters.m_strategyName = "linear"; // can be "pca" instead to use principal components analysis
 
// link specific parameters with chosen implementation
te::rp::MixtureModelLinearStrategy::Parameters lmmParameters;
mmInputParameters.setMixtureModelStrategyParams(lmmParameters);
 
// defining endmembers
mmInputParameters.m_components["clouds"].push_back(255.0);
mmInputParameters.m_components["clouds"].push_back(184.875);
mmInputParameters.m_components["clouds"].push_back(255.0);
 
mmInputParameters.m_components["shadow"].push_back(42.075);
mmInputParameters.m_components["shadow"].push_back(24.99);
mmInputParameters.m_components["shadow"].push_back(49.98);
 
mmInputParameters.m_components["vegetation"].push_back(99.96);
mmInputParameters.m_components["vegetation"].push_back(64.005);
mmInputParameters.m_components["vegetation"].push_back(154.02);
 
// defining sensor information
mmInputParameters.m_inputSensorBands.push_back("CBERS2_CCD_3_RED");
mmInputParameters.m_inputSensorBands.push_back("CBERS2_CCD_4_NIR");
mmInputParameters.m_inputSensorBands.push_back("CBERS2_CCD_2_GREEN");
 
// defining output parameters
std::map<std::string, std::string> orinfo;
orinfo["URI"] = "output-linear-mm.tif";
mmOutputParameters.m_rInfo = orinfo;
mmOutputParameters.m_rType = "GDAL";
mmOutputParameters.m_normalizeOutput = true;
mmOutputParameters.m_createErrorRaster = true;
 
// execute the algorithm
te::rp::MixtureModel mmInstance;
bool initok = mmInstance.initialize(mmInputParameters);
if (initok)
  mmInstance.execute(mmOutputParameters);
 
// clean up
delete rin;

Dependencies: Raster