Application of Neural Network Technologies for the Classification of Cloudiness by Texture Parameters of MODIS High-Resolution Images

Abstract

The technique of a search for images of cloudiness of various types from MODIS satellite images based on a comparison with archive data of observations on the network of meteorological stations is presented. Based on an expert estimate, 14 types of cloudiness possessing a unique structure on images recorded with a spatial resolution of 250 m are identified. Images of cloudiness of these types and results of investigations of their texture parameters found based on the statistical gray-level co-occurrences matrix (GLCM) approach are presented. For the indicated cloudiness types, characteristic texture features or their combinations are determined. To classify the cloudiness based on information on the texture parameters, it is proposed to use the neural network based on the three-layer perceptron. The modified method of adaptive tuning of the learning rate of the neural network is described. Results of cloudiness classification and their reliability are discussed.