A technique for testing hypotheses for distributions of multidimensional spectral data using a nonparametric pattern recognition algorithm

The paper deals with a new method of testing hypotheses for the distribution of multidimensional remote sensing spectral data. The proposed technique is based on the use of nonparametric algorithms for pattern recognition. Testing the hypothesis of the identity of two laws of distributions of multidimensional random variables is replaced by testing a hypothesis stating that the pattern recognition error equals 0.5. The application of this technique allows doing without the decomposition of the random variable domain into multidimensional intervals, which is typical for the Pearson criterion. Its effectiveness is confirmed by the results of testing the hypotheses of the distribution of spectral data of remote sensing in forestry. The analysis of the distribution laws for the following types of forestry is carried out: dark coniferous forest, damaged and dry forest stands. The initial information was obtained from the southern Siberia remote sensing data using six spectral channels of Landsat. The results of the research form a basis for a set of significant spectral features when dealing with forest condition monitoring.