ABSTRACT
This chapter considers three representative models for supervised classification which involve this sort of probability density estimation: a parametric model, a nonparametric model, and a semiparametric or mixture model. Land cover classification of remote sensing imagery is a task which falls into the general category of pattern recognition. Pattern recognition problems, in turn, are usually approached by developing appropriate machine learning algorithms. Support vector machine are also nonparametric in the sense that they make direct use of a subset of the labeled training data to effect a partitioning of the feature space, however, unlike the aforementioned classifiers, without reference to the statistical distributions of the training data. The renaissance of interest in neural networks is due to the success of so-called deep learning algorithms for the application of machine learning to very large data sets. In essence one works with neural networks of varying architectures, all characterized by having many hidden layers and correspondingly many free parameters.
