DIFFERENTIAL CLASSIFICATION WITH MULTIPLE NEURAL NETWORKS Russian patent published in 2018 - IPC G06N3/08 G06F17/18 G06K9/66 

FIELD: data processing.

SUBSTANCE: invention relates to means of text analysis. Kernel of the classification system stores in several neural networks memory, each is trained to recognize a set of one or more sets of confusing graphemes defined in the recognition data of a plurality of document images. Input image of the grapheme, associated with the image of the document, containing a lot of graphemes, turns out. Variety of variants for recognizing an input image of a grapheme are determined, where a plurality of recognition options includes a plurality of target symbols that are similar to an input grapheme image are defined. First neural network are selected of a plurality of neural networks, the first neural network is trained to recognize the first set of entangled graphemes, and wherein the first plurality of graphemes contains at least a portion of a plurality of recognition patterns for the input image of the grapheme. Grapheme class for the input image of the grapheme using the selected first neural network is defined.

EFFECT: technical result consists in reducing the amount of computing resources while recognizing the text.

20 cl, 8 dwg