Pixel allocation using remotely-sensed data and ground data
Two models for integrating information from different sources with the purpose of classifying pixels are proposed. In particular we have in mind situations where remotely-sensed spectral data and ancillary ground data are available for each pixel in a given area. The issues addressed are to find models which integrate these two sources of data, and to investigate to what extent the local uniformity of the ground data captures the spatial correlation. The label of each pixel is unobserved and hence an EM algorithm is used for estimating the relevant probabilities. Experiments based on real data are performed.