Neural network classifier trained for purchasing differentiation

Patent metadata

Field	Value
Publication number	US-11651664-B2
Application number	US-202117171143-A
Country	US
Kind code	B2
Filing date	Feb 9, 2021
Priority date	Feb 10, 2020
Publication date	May 16, 2023
Grant date	May 16, 2023

Abstract

Systems and methods for self-checkout at a point-of-sale are provided. The system and method includes using a plurality of radio frequency identification (RFID) transceivers within a store, and an RFID reader configured to receive an RFID code from an RFID tag activated by the plurality of radio frequency identification (RFID) transceivers. The system and method also includes using a classifier configured to determine whether the RFID tag is inside or outside a designated area, wherein the classifier is trained in a manner that a number of items incorrectly identified as being purchased is below a threshold to minimize customer dissatisfaction (CDS) determined as the ratio of the value of items charged to the customer but not purchased by a customer to the total charge to the customer.

First claim

What is claimed is:

1. A system for self-checkout at a point-of-sale, comprising:
a plurality of radio frequency identification (RFID) transceivers within a store;
an RFID reader configured to receive an RFID code from an RFID tag activated by the plurality of radio frequency identification (RFID) transceivers;
a pair of RF-absorbing walls;
entrance sensors at an entrance side of the pair of RF-absorbing walls configured to determine that a person has entered a checkout area;
exit sensors at an exit side of the pair of RF-absorbing walls configured to determine that a person has exited the checkout area;
a billing terminal with a payment processor configured to receive payment information from the customer in the checkout area; and
a multilayer neural network classifier configured to determine whether the RFID tag is inside or outside the checkout area, wherein the multilayer neural network classifier is trained in a manner that a number of items incorrectly identified as being purchased is below a threshold to minimize customer dissatisfaction (CDS) determined as the ratio of the value of items charged to the customer but not purchased by a customer to the total charge to the customer, and wherein the multilayer neural network classifier minimizes a combination of a vendor loss (VL) determined as a ratio of the value of items that exit the store without being charged to the customer to the total amount of the value of the items taken from the store and paid for plus the value of the items charged to the customer, but not actually purchased, and the customer dissatisfaction (CDS).

2. The system as recited in claim 1 , wherein the charge to the customer is adjusted after the customer has exited the store.

3. The system as recited in claim 2 , wherein the adjustment to the charge is determined by a subsequent inventory of store items, and wherein vendor loss is given by: VL=f(FN)/(f(TP)+f(FP)),
where f(FN) is the value of the items in the set of False Negatives for items taken from the store, but not identified by the classifier as included in set P, f(TP) is value of the items in the set of True Positives for items taken from the store and paid for, and f(FP) is value of the items in the set of False Positives for items charged to the customer, but not actually purchased.

4. The system as recited in claim 2 , wherein the store checkout area (CA) is determined by the signal coverage of the plurality of radio frequency identification (RFID) transceivers and the pair of RF-absorbing walls separated by a distance.

5. The system as recited in claim 4 , further comprising a waiting area (WA) adjoining the checkout area, where at least a subset of the plurality of radio frequency identification (RFID) transceivers are oriented to send signals to and receive signals from the waiting area (WA).

6. The system as recited in claim 5 , wherein the RFID reader initiates a checkout process by sending an energizing signal from the plurality of radio frequency identification (RFID) transceivers.

7. A system for self-checkout at a point-of-sale, comprising:
a pair of RF-absorbing walls separated by a distance;
entrance sensors at an entrance side of the pair of RF-absorbing walls configured to determine that a person has entered a checkout area;
exit sensors at an exit side of the pair of RF-absorbing walls configured to determine that a person has exited the checkout area;
a billing terminal with a payment processor configured to receive payment information from the customer in the checkout area;
a plurality of radio frequency identification (RFID) elevated transceivers within each of the pair of RF-absorbing walls;
a plurality of floor-mounted transceivers in a floor below the pair of RF-absorbing walls, wherein the signal coverage of the plurality of radio frequency identification (RFID) transceivers determines a checkout area (CA);
an RFID reader, wherein the RFID reader is configured to receive an RFID code from each of a plurality of RFID tags activated simultaneously by the plurality of radio frequency identification (RFID) transceivers, the RFID reader configured to receive each of the plurality of RFID codes in a separate time slot of a block of a transmit frame; and
a classifier configured to determine whether the RFID tag is inside or outside a designated area, wherein the classifier is trained in a manner that a number of items incorrectly identified as being purchased is below a threshold to minimize customer dissatisfaction (CDS) determined as the ratio of the value of items charged to the customer but not purchased by a customer to the total charge to the customer.

8. The system as recited in claim 7 , wherein the classifier is a neural network model that applies weights to an input feature vector and outputs a probability for each RFID tag being inside the designated area.

9. The system as recited in claim 8 , further comprising a waiting area (WA) adjoining the checkout area, where at least a subset of the plurality of radio frequency identification (RFID) transceivers are oriented to send signals to and receive signals from the waiting area (WA).

10. The system as recited in claim 9 , wherein the classifier minimizes a combination of a vendor loss (VL) determined as a ratio of the value of items that exit the store without being charged to the customer to the total amount of the value of the items taken from the store and paid for plus the value of the items charged to the customer, but not actually purchased, and the customer dissatisfaction (CDS), where a classifier algorithm partitions a set of objects into positive and negative subsets, wherein the set of objects includes all items detected by the system for self-checkout at the point-of-sale based on a comparison of a metric function value for each items and a threshold.

11. The system as recited in claim 10 , wherein the charge to the customer is adjusted after the customer has exited the store.

12. A method for self-checkout at a point-of-sale, comprising:
detecting that a person has entered a checkout area defined by a pair of RF-absorbing walls, entrance sensors at an entrance side of the pair of RF-absorbing walls configured to determine that a person has entered the checkout area, and exit sensors at an exit side of the pair of RF-absorbing walls configured to determine that a person has exited the checkout area, wherein activation of the entrance sensors trigger an energizing signal;
determining whether an RFID tag is inside the checkout area (CA) using a multilayer neural network classifier and a plurality of radio frequency identification (RFID) transceivers, wherein the multilayer neural network classifier minimizes a combination of a vendor loss (VL) determined as a ratio of the value of items that exit the store without being charged to the customer to the total amount of the value of the items taken from the store and paid for plus the value of the items charged to the customer, but not actually purchased, and the customer dissatisfaction (CDS) determined as the ratio of the value of items charged to the customer but not purchased by a customer to the total charge to the customer, where a classifier algorithm partitions a set of objects into positive and negative subsets, wherein the set of objects includes all items detected by the system for self-checkout at the point-of-sale based on a comparison of a metric function value for each items and a threshold; and
charging the person an amount for items associated with the RFID tag determined to be inside the checkout area.

13. The method as recited in claim 12 , further comprising determining that the person has exited a checkout are

Assignees

• Nec Lab America Inc
• Nec Corp

Inventors

• Khojastepour Mohammad
• Arslan Mustafa
• Rangarajan Sampath
• Chai Eugene
• Bocanegra Carlos

Classifications

• G06N3/0499

  Feedforward networks · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06Q20/204

  comprising interface for record bearing medium or carrier for electronic funds transfer or payment credit · CPC title

• G06N3/08

  Learning methods · CPC title

• G06Q10/087 Primary

  Inventory or stock management, e.g. order filling, procurement or balancing against orders · CPC title