Numerical methods of structure-statistical object classification for remote sensing

The main stage of solution of the structure-statistical object classification problem for remote sensing is the construction of probability distribution law on the base of probability learning sample. The structure of the solution of the problem includes the following five problems: the algorithm of construction of initial approximation of iterative process of estimation of mixed model parameters; the iterative process of EM-algorithm for estimation of distribution law parameters; criterion of the end of iterative EM-algorithm process; the determination of the number of components of mixed model of object feature-characteristic law distribution; statistical classification of objects of probability learning sample (PLS). The solution of each problem requires the building of new mathematical models, the creation of a number of simulation examples and computer programs, the improvement of existent algorithms for minimization of calculation time, and the check of program software on real computer data.