METHOD OF TRAINING GOODS RECOGNITION SYSTEM ON IMAGES Russian patent published in 2019 - IPC G06Q30/00 

FIELD: information technology.

SUBSTANCE: invention relates to systems for recognition of goods on images. Method comprises steps of extracting from a data storage unit an image of goods obtained using user devices using a neural network training apparatus; forming, for each commodity image, a multidimensional image vector; clustering formed multidimensional image vectors to obtain image clusters; identifying at least one cluster of images either as one of a known system of goods, or as a new previously unknown commodity; assigning to at least one cluster of images, which has been identified as a new commodity, a unique cluster label; teaching a neural network to recognize goods on images based on images of said image clusters, wherein the commodity, which was previously determined as an unknown commodity, is assigned a unique cluster mark as a unique commodity characteristic; forming, in the catalog of goods, a new type of commodity, which as a unique characteristic of the commodity is assigned a unique cluster mark, and examples of images of the goods are assigned images which entered the cluster with this unique label of the cluster.

EFFECT: automated training of neural network based on new images of goods.

4 cl, 1 dwg