Revision as of 13:13, 18 June 2021

Using QGIS and OTB processing toolbox

• Using OTB version > 7.2.0, input vector file format of training data can be GeoPackage GPKG or ESRI Shapefile.
• The column containing class label needs to be an integer (whole number).

Training phase

• In the search engine of Processing Toolbox, type TrainImages and open TrainImagesClassifer.
• In the Input Image List select one (or optional: several) multi-band raster (multispectral images).
• In the Input Vector Data List do not select a vector file from the list of files which might be already loaded in the QGIS Viewer. There is currently a bug in QGIS 3.16 which leads to failure during file import. Instead please select a vector file clicking and browse directly to the file containing training area polygons in format GPKG or SHP.
• If you do not have independent validation data: leave Validation Vector Data List empty.
• Bound sample number ba minimum: 1. The class with the minimum number of pixels determines the sample size of all other classes. Changing this value to 0 does not have an effect (current bug in QGIS3.16).
• Training and validation sample ratio: 0.25. This is a 4-fold cross-validation with split 0.75 per cent for training and 0.25 per cent for testing.
• Type C_ID in the Field containing the class integer label for supervision text field.
• As Classifier to use for training choose Support Vector Machine libsvm from the drop down list.
• The SVM Kernel Type is Linear.
• SVM Model Type is csvc
• Switch checkbox Parameters optimization off. The optimization results in a higher accuracy but takes much time (> 1 hour computation).
• In the Output model specify a model file: e.g. svm.model
• Click Run.
• Click on the Log tab and inspect the model quality measures: Precision, Recall, F-score and Kappa index.

Classification phase

• In the search engine of Processing Toolbox, type ImageClassifier and double click ImageClassifier.
• Set an multiband image as Input image.
• Set Input _mask to blank (top of drop-down list).
• Set svm.model as Model file.
• Set the number of classes in your model: this is the number of unique classes in your training vector file.
• Set Output pixel type to uint8
• Save the Output image as svm_classification.tif.
• Uncheck Confidence map: Open output file after running algorithm.
• Uncheck Probability map: Open output file after running algorithm. Run.

• Find the output svm_classification.tif in the QGIS map canvas.
• Right click on the layer svm_classification and select Properties --> Symbology --> Style --> Load Style.
• Select the style file classified.qml. OK.

Postprocessing

This application uses a majority filter with a circular structure element.

• In the search engine of Processing Toolbox, type regularization and double click ClassificationMapRegularization.
• Define Input classification image as an classification output raster.
• Set Structuring element radius to 2 pixels.
• Set Output pixel type to uint8.
• Save the Output regularizd image as svm_classification_majority.tif.
• Click Run.

• Find the output svm_classification_majority.tif in the QGIS map canvas.
• Right click on the layer svm_classification_majority in the and select Properties --> Symbology --> Style --> Load Style.
• Select the style file classified.qml. OK.

Using OTB standalone

Training phase

• Navigate in the Windows explorer to your Folder OTB-7.2.0-Win64. Double click the Windows-Batchdatei mapla to open Monteverdi Application Launcher.
• In the search engine of mapla, type TrainImages and double click TrainImagesClassifer.
• In the Input Image List click on + and select a (or optional: several) multispectral images: Subset_S2A_MSIL2A_20170619T_MUL.tif .
• In the Input Vector Data List choose a vector polygon file with training areas in file format GPKG or SHP.
• In the Output model specify an output model file: e.g. lucc_rf.model
• In the Bound sample number by minimum field type 0.
• Set the training and validation sample ratio to 0.25.
• Mark C_ID in the Field containing the class integer label (C_ID refers to the column that contains the LUC code in the training and validation vector file).
• Choose Shark Random forests classifier from the drop down list as Classifier to use for the training.
• Check user defined Random seed and enter any positive integer value. This initializes a pseudorandom number generator which may be used to reproduce results.
• Click on Execute.
• Click on the Log tab and inspect the model quality measures: Precision, Recall, F-score and Kappa index.

Classification phase

• In the search engine of mapla, type ImageClassifier and double click ImageClassifier
• Set Subset_S2A_MSIL2A_20170619T_MUL.tif as Input image.
• Set lucc_rf.model as Model file.
• Save the Output image as rf_classification.tif.
• Adjust the Number of classes in the model to the number of unique classes in the training vector file.

• Add rf_classification.tif to QGIS canvas.
• Download the style file classified.qml from Stud.IP.
• Right click on the layer rf_classification and select Properties --> Style --> Style --> Load Style.
• Select the style file classified.qml. OK.

Postprocessing

This application uses a majority with a circular structure element.

• In the search engine of mapla, type, regular and double click ClassifcationMapRegularization.
• As theInput classification image define the output raster file of the classification phase.
• Save the Output regularizd image as rf_classification_majority.tif.
• Set Structuring element radius to 2 pixels.
• Set Output pixel type to uint8.
• Click Execute.

• Add rf_classification_majority.tif to QGIS canvas.
• Download the style file classified.qml from Stud.IP.
• Right click on the layer rf_classification_majority and select Properties --> Style --> Style --> Load Style.
• Select the style file classified.qml. OK.