Signaturen im zweidimensionalen Merkmalsraum
Unsupervised Classification – k-MEANS 2-D Scatter Plot Band 2 Band 1
Unsupervised Classification – k-MEANS Initial class means – evenly distributed Parameters: (a) number of classes (b) distance threshold Band 2 Band 1
Unsupervised Classification – k-MEANS Classification by minimum distance Band 2 Band 1
Unsupervised Classification – k-MEANS Classification by minimum distance (first iteration) Band 2 Band 1
Unsupervised Classification – k-MEANS Recalculation of new means (cluster centres) Band 2 Band 1
Unsupervised Classification – k-MEANS Second iteration Band 2 Band 1
Unsupervised Classification – k-MEANS Stop criteria: (a) maximum iteration (b) number of changed pixels Band 2 Band 1
Unsupervised Classification Klassen werden selbständig anhand der Bildstatistik gebildet Automatische Zusammenfassung von Pixeln zu Clustern Cluster = Klasse Gefundene Klassen müssen nach der Klassifizierung bestimmt werden Anwendung, wenn das Gebiet und die Materialien unbekannt sind Rechenaufwand hoch Aufwand für Bearbeiter gering
Unsupervised Classification – k-MEANS - Summary Randomly generated initial cluster centres Each pixel is associated with the nearest cluster centre Recalculation of the cluster centres Steps 2 and 3 are repeated until the given number of iterations or until a given number of pixels isn’t changing any more 4 Cluster 5 Cluster 6 Cluster