Numerical method for predictive estimate of classification efficiency for various cloud type images based on texture information from MODIS data

摘要

A numerical method is proposed for constructing successive schemes for making decisions about the attribution of images of various cloud types to certain classes, which makes it possible to form sets of informative features and to evaluate their efficiency for subsequent classification. For formalized description of images, statistical methods of texture analysis are used: grey-level co-occurrence matrices, grey-level difference vectors, sum and difference histograms of image brightness levels, and statistical characteristics of individual pixel brightness. The results are discussed of the developed method testing for classification of 25 cloud types on MODIS images according to the current standard of the World Meteorological Organization. Recommendations are given for predictive estimate of classification efficiency of cloud cover images.