Self-Checkout Picklists for Organic Produce Misclassification

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing machine learning models struggle to accurately predict visually indistinguishable produce items at self-checkout, leading to potential profit loss for retailers and degraded consumer experience due to incorrect item recommendations.

Innovation Solution

A machine learning model configured to output organic equivalents alongside non-organic produce items in the picklist, even if the organic items do not meet the confidence threshold, and adjusts feedback data to enhance accuracy by treating organic selections as non-organic, ensuring accurate sales capture without lowering the confidence threshold.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If the confidence threshold is lowered to include organic produce items in the picklist, then the completeness of item predictions is improved, but the precision of predictions deteriorates

Engineering Contradiction:

Improvecompleteness of item predictionsVSAvoidprecision of predictions

Core Design Contradiction:

ReliabilityVSMeasurement precision

Solution Approach 1:

The patent segments the picklist generation process into two distinct stages: (1) generating candidate predictions using the machine learning model with the existing confidence threshold, and (2) supplementing the picklist by adding organic equivalents for non-organic items that appear in the predictions. This segmentation allows the system to maintain the original precision threshold while still achieving completeness by adding organic items through a separate supplementation mechanism rather than lowering the threshold for all items.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The system performs preliminary identification of non-organic produce items in the picklist predictions, then proactively adds their organic equivalents before presenting the final picklist to the consumer. This preliminary action ensures organic items are included without requiring the confidence threshold to be lowered, as the organic items are added based on the presence of their non-organic counterparts rather than independent confidence scoring.

Inventive Principle:

Principle #10Preliminary action

2Reliability

If organic item PLU codes are returned to the machine learning model for continuous learning, then the model's ability to predict organic items is improved, but the accuracy of non-organic item predictions deteriorates

Engineering Contradiction:

Improveability to predict organic itemsVSAvoidaccuracy of non-organic item predictions

Core Design Contradiction:

ReliabilityVSMeasurement precision

Solution Approach 1:

Instead of feeding organic item PLU codes back to the machine learning model for training, the patent inverts the approach by feeding non-organic item PLU codes back to the model. This inversion allows the model to learn from actual consumer selections of non-organic items, improving its ability to predict non-organic produce while maintaining the separate mechanism for identifying and adding organic equivalents through PLU code supplementation.

Inventive Principle:

Principle #13The other way round (Inversion)

3Adaptability or versatility

If the picklist is supplemented to include organic equivalents, then consumer choice and accuracy are improved, but the device complexity increases

Engineering Contradiction:

Improveconsumer choiceVSAvoidpicklist generation complexity

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent introduces an intermediary supplementation step that acts as a bridge between the machine learning model's predictions and the final picklist presentation. This intermediary process automatically identifies non-organic items in the predictions, retrieves their organic equivalent PLU codes from a database, and inserts them into the picklist in appropriate positions. This intermediary mechanism adds functionality without requiring fundamental changes to the machine learning model or the overall system architecture.

Inventive Principle:

Principle #24Intermediary (Mediator)

Data Source

PatentUS20260038022A1Techniques for Enhancing the Relevancy of Candidate Item Selections Presented at a Self-Checkout

Publication Date: 2026.02.05 NCR VOYIX CORP

AI summary

A picklist of candidate predicted items for a non-barcoded item being purchased at a self-checkout terminal is enhanced to include other candidate predicted items that are visually indistinguishable from the candidate predicted items. An organic produce item PLU is included in the picklist for each non-organic PLU in the picklist. The organic produce item PLUs are ordered adjacent to the corresponding non-organic item PLUs to improve the consumer browsing experience and increase the likelihood that organic item purchases are accurately captured. Consumer selections from the picklist are returned to a machine learning model as feedback data to enable continuous learning and improved model accuracy. Non-organic produce item PLUs are returned for selections of organic produce items, thereby resulting in improved confidence values for non-organic PLUs and increased prediction accuracy of the model, while still ensuring organic items are included in the picklist and without requiring lowering of a confidence threshold.