5-7 Classification Methods

Computer assisted classification of multispectral imagery in remote sensing is useful for thematic mapping of land use, vegetation, soil, geology etc.

Classification methods are classfied into two categories.

Supervised classification: classification with the use of ground truth data in the form of sample sets. Maximum likelihood classifier is one of the typical supervised classification methods.

Unsupervised classification: classification with only spectral features without use of ground truth data. Clustering is an unsupervised classification in which a group of the spectral values will regrouped into a few clusters with spectral similarity.

The following classification methods are widely used depending on the spectral characteristics and availability of ground truth data as shown in Figure 5.10.

Rationing: classification between vegetation and non-vegetation is possible.

Box classifier: very easy to apply level slicing but accuracy is not very high.

Discriminant function: useful in the case when the number of classes is not many.

Clustering: unsupervised classifier with spectral similarity or distance between clusters.

Minimum distance method: several statistical distance measures such as Euclidan, Mahalanobis, Bhattacharya and Jefrey-Matsushita distance are used to determine the class.

Maximum likelihood classifier: see 5-8
Knowledge-based classification including decision tree classifier will be specified as classification model by users.