| CPC G06Q 30/018 (2013.01) [G06K 7/1413 (2013.01); G06N 3/08 (2013.01); G06N 5/04 (2013.01); G06N 7/01 (2023.01); G06Q 20/18 (2013.01); G06Q 20/20 (2013.01)] | 15 Claims |
|
1. A method, comprising:
receiving, by a processor of a server, item images of an item captured by cameras during a transaction at a transaction terminal, wherein receiving further includes receiving at least one of the item images from a certain overhead camera of the cameras, wherein the certain overhead camera is focused down on a top surface of a weigh scale and provides the at least one of the item images when an item is placed on the weigh scale during the transaction;
producing, by the processor, a feature vector based on the item images, image captured of an item by cameras during a transaction at a transaction terminal, wherein at least one item image captured by an overhead camera of a weigh scale for the transaction terminal with the item placed on the weigh scale during the transaction, wherein producing further includes processing the item images using a trained Convolutional Neural Network (CNN) model to generate the feature vector, wherein the trained CNN model is configured to output a low-dimensionally, discriminative feature vector for produce items, and wherein the trained CNN model and Bayesian produce recognition engines are leveraged across a plurality of disparate retailers and a plurality of stores associated with each retailer for purposes of identifying and verifying the produce items during transactions at terminals;
wherein the Bayesian produce recognition engines each utilize an equation that sums elements of the feature vector and sales data for each Produce Look Up (PLU) code to produce a probability that the feature vector and sales data for a corresponding image of a particular produce item is or is not the PLU code associated with a corresponding Bayesian produce recognition engine;
weighting, by the processor, the feature vector based on transaction history data for an entered item code received from the transaction terminal for the item or for available item codes that are available to the transaction terminal; and
providing, by the processor, a list of candidate item codes based on the feature vector and the weighting and iterating back to producing the feature vector when a specific item code that is not included in the list of candidate item codes is provided by an operator of the transaction terminal or providing a verification for the item based on the entered item code or the specific item code selected by the operator, the feature vector, and the weighting.
|