Transcript
Copyright © Department of Geoinformatics, Gdansk University of Technology
Laboratory of Geographic Information Systems Multispectral image classification in Quantum GIS Classification is a process that may assist in vectorization and interpretation of a raster image. The name image classification refers to the procedure by which the individual pixels of the image are assigned to specific groups (classes), through assigning each pixel a specific number which represents the class to which the pixel was assigned. The principle of selecting the class to which each pixel is assigned, in general, can be described as follows: a single class should include pixels with similar values, and pixels with considerably different values should end up in different classes. At the same time, the number of classes should not be too large. In satellite images these classes usually represent specific types of terrain, such as vegetation, water, built-up areas, rocks, etc.. This is because a particular type of terrain usually radiates with a certain intensity in different frequency ranges recorded by the satellite. Therefore a classification of such an image is very likely to reveal such areas by assigning them to different classes. Such an image can then be subject to a raster-vector conversion, in which areas of each class will be converted into (generally multipart) polygons. TASKS TO BE CARRIED OUT 1. Perform supervised classification of the given satellite image.
(up to) 5 pt.
INSTRUCTIONS Begin with starting Quantum GIS and enabling the image classification module. For this purpose go to Plugins->Manage and Install Plugins and click the Get more section on the panel to your left. From this list select Semi-Automatic Classification and click Install plugin. You can now open the satellite image for analysis through Layer->Add Raster Layer. Once the image has been opened, it must be associated with the classification tools. For this purpose, in the classification toolbar (top-side of the window), select it as "Input image". In the same toolbar, set the “RGB=” value to "3-2-1". To begin classification, we must first specify Training Areas. Training Areas are regions of the input image which are representative of certain feature classes. To create Training Areas, go to "SCP:ROI creation" (right-side panel) and click "New shp" under "Training shapefile". Save this new shapefile in your directory. Now, find 5-6 different feature classes in the image (such as grass, forests, buildings, water bodies, etc.). For every class create a Training Area (ROI) by clicking the orange polygon icon under "ROI creation" (right-side panel) and drawing the area on the map (draw by leftclicking, finish drawing with a right-click). Make sure that the selected areas are uniform! If you have problems distinguishing between eg. different types of vegetation, change the "RGB=" value in the classification toolbar to "4-3-2". Once you have drawn a new ROI, give it a new Macroclass ID (starting from "1") and Macroclass Name (eg. "Buildings" or "Vegetation") as well as Class ID (also starting from "1") and Class Name (eg. "Airport building" or "Forest Trees") under "ROI Signature definition" (right-side panel). Remember that Multiple Class IDs can be assigned to the same Macroclass ID, but the same Class ID cannot be assigned to multiple Macroclass IDs ! (ie. Macroclass "Vegetation" can have an ID of "1", and its members "Grass" and "Trees" can have Class IDs "1" and "2", respectively. However, the first member of Macroclass ID "2", named "Buildings" must have a Class ID of "3", because "1" and "2" are already taken by members of Macroclass "1"). After assigning proper values to all required fields, click "Save ROI". 1
Copyright © Department of Geoinformatics, Gdansk University of Technology
When Training Areas for all classes have been created, give them appropriate colours under "Signature list" (left-side panel). Once the classes have been specified and coloured, select "Minimum Distance" under "Classification algorithm" (left-side panel). You can now perform a preview of the classification results by clicking on the "+" button under "Classification preview" (left-side panel) and clicking on the desired spot on the map. If you are not happy with the results, you can always correct your Training Areas or add new ones (just make sure to assign appropriate Class and Macroclass IDs). When you are ready, click "Perform Classification" and save the result to your directory.
2
