METHOD OF CLASSIFICATION OF COMPLEX STRUCTURED IMAGES ON THE BASIS OF SELF-ORGANIZED NEURAL NETWORK STRUCTURES

The relevance in development of intelligent systems for classification of complex structured images occurs during processing of images taken from cameras of UAV used for navigational purposes where there is no communication with artificial Earth satellites, or in the analysis of images in real time by the operator. The developed method provides high requirements to quality of classification of objects on images, as well as to fast selection and classification of investigated segments of images. For classification of such images the appropriate computer technologies based on the boosting methodology are offered. The space of Informative features is formed by spectral windows obtained by scanning of the original image. Spectral windows belonging to different classes, are arranged in the form of clusters on Kohonen plane. To form a cluster, the rules of correction of vectors of weights are used, and such rules make it possible to reduce the values of insignificant components of the vectors and the coordinates of the clusters centers are identified. Strong classifiers are built on the basis of the cluster structure of Kohonen plane. There has been designed and demonstrated the structure of the strong classifier on neural networks of direct distribution referred to block type, which was implemented for classification of chest X-ray images.