A process to categorize all pixels in a digital image into one or several land cover classes or themes. It uses the spectral information represented by the digital numbers in one or more spectral bands and classify each individual pixel based on this spectral information(DN). Based on DN values we can create the classes like vegetation, water body, barren land, build up area etc.
Differentiate between Supervised and unsupervised classification (three distinctions)
| Supervised classification |
Unsupervised classification |
| a) The analyst recognizes classes in any image based on prior knowledge. These are known as training sites. |
a) Spectral classes are grouped first based on the numerical information in the data. It does not utilize training data. |
| b) It frequently uses parallelepiped, minimum distance and maximum likelihood algorithms |
b) The most used algoristhm is “K means” approach also called ISODATA |
| c) It is more accurate for mapping classes. |
c) It is not complete without human involvement. |
