A modified method of structural analysis of remote sensing data: normalization of spectral feature values by distribution mode

The problem of increasing the efficiency of the method of structural analysis of remote sensing data by using a modified estimate of the correlation coefficient is considered. The traditional method of structural analysis of remote sensing data is based on the normalization of spectral features by their average values and does not take into account the features of asymmetric distribution laws of random variables. The proposed method is based on the analysis of the components of the modified estimation of the correlation coefficient of spectral features, which are normalized according to the modes of their distributions. The number of components in the modified correlation coefficient estimate corresponds to the volume of the original data. Based on this, a decision rule is formed that identifies four classes. The detected classes correspond to the component values of the modified correlation coefficient estimate, which take positive, alternating, and negative values. Nonparametric probability density estimates are used to identify the modes of the distribution laws for spectral feature values. A method based on maximizing the likelihood function was used to optimize the nonparametric probability density estimate based on the blur coefficients. An alternative method for optimizing the kernel probability density estimate is based on minimizing the estimates of their standard deviations. To compare traditional and modified methods of structural analysis of statistical data, a computational experiment was conducted to process remote sensing data from a test area of forest cover damaged by the Siberian silkmoth. Interpretation maps of the forest test area are analyzed using spatial coordinates corresponding to traditional and modified methods of structural analysis of remote sensing data. The numerical characteristics of the obtained results from the computational experiment are compared and analyzed. It was found that the area of damaged forest cover determined using the modified structural analysis method is more consistent with the results of fi eld studies of the test site compared to the traditional method. The obtained results are relevant for the development of information systems for deciphering space imagery materials.