How do you know the type of land use each spectral represents from an remote sensed data into a classification ?
Which classification is the most typical to use when doing that ?
How many classification are there do you recommend ?
Geographic Information Systems Stack Exchange is a question and answer site for cartographers, geographers and GIS professionals. It only takes a minute to sign up.Sign up to join this community
@nicksan is right. In its simplest form, spectral (pixelwise) classification is based on a set of examples that you define manually. In most RS-oriented software (ENVI, ERDAS, Orfeo Toolbox, etc), this is based on photointepretation, i.e. drawing polygons or sampling pixel of each ground cover class, modeling them with some learning technique and generalizing the learned rules on the rest of the pixels composing the image.
You will need a sufficient number of pixels for each class, which is usually proportional to the number of spectral classes you are interested in and the dimensionality of the image (e.g. think to RGB vs hyperspectral images).
Most RS software usually offer simple and off the shelf classification algorithms, ranging from simple Gaussian maximum likelihood to more complex support vector machines and neural networks. In any case, the software guide itself should help.
The type of classifier to be employed (parametric, nonparametric, etc) usually depends on which kind of land cover classes are you trying to discriminate. In general, nonparametric classifiers reduce the assumptions you should make about class distribution. However, depending on the problem this observation may vary quite a lot. To have an idea of existing techniques, just visit for instance remote sensing related journals and look for image classification and semantic segmentation.