Algorithms of object identification on the basis of hyperspectral Earth survey data using fuzzy linear regression

An approach to solving problems of identifying Earth surface objects on the basis of hyperspectral survey data obtained from space complexes based on the comparison of hyperspectral characteristics of the objects investigated with a set of reference signatures is presented in the paper. Algorithms of object identification with the use of the theory of fuzzy sets are proposed: an identification algorithm based on fuzzy linear regression and an algorithm of consolidation of results of different identification solutions. The fuzzy linear regression algorithm is based on the use of non-symmetrical triangular fuzzy numbers. This approach, used earlier in solving approximation tasks and assessing the unique character of electronic map fragments is now used for the first time for the identification of hyperspectral characteristics. The choice is founded on the fact that fuzzy linear regression makes identification possible in ambiguous conditions. The results of experimental studies of the proposed algorithms based on real hyperpsectral survey data (from spacecraft «Resurs-P» N1) are presented in the form of 10 images. Identification reliability is shown to increase by 6.1 % as compared with one of the initial algorithms yielding the best solution in the case of consolidating results obtained by using algorithms based on the Euclidean distance similarity measure, the angle similarity measure, as well as fuzzy similarity measures.