Utilizing Multitemporal Data by a Stochastic Model

In remote sensing, a principle objective is to produce an accurate ground cover thematic classification map. In this paper a new classifier which makes use of multitemporal data is described. Ground cover types are considered as stochastic systems with nonstationary Gaussian processes as input and temporal variation of reflected and emitted electromagnetic energy as output. Then, by assumption that the behavior of these stochastic systems are governed by first-order Markov processes, multitemporal information is utilized. As a result of this approach for characterizing multitemporal data, a new processor, the Markov classifier, is developed. Experimental results from Landsat MSS data are included.