Article Image Classification Using Availability-Weighted Inference

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing article discrimination systems inaccurately infer the type of an article when it is not available, such as being out of stock, due to similarities in image patterns.

Innovation Solution

An article discrimination system utilizing machine learning to train a discriminator to infer the type of an article based on captured images, with a mechanism to lower the probability of inferring unavailable articles, ensuring they are not completely excluded from candidates, and incorporating availability information from store databases.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If the article discrimination system uses image pattern recognition to infer article types, then the inference speed is fast, but the accuracy deteriorates when articles are unavailable (out of stock) due to similar image patterns causing misclassification

Engineering Contradiction:
Improvearticle type inference accuracyVSAvoidinference reliability for unavailable articles
Core Design Contradiction:
Measurement precisionVSReliability

Solution Approach 1:

The patent introduces availability information as an intermediary element that mediates between the image recognition result and the final article type determination. When an article is identified through image recognition, the system checks its availability status before finalizing the inference. This intermediary check prevents misclassification of unavailable articles while maintaining the speed of image-based recognition.

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The system dynamically changes the inference parameters based on article availability status. For available articles, the system uses standard image recognition confidence thresholds. For unavailable articles, the system adjusts the probability thresholds and weighting factors to reduce the likelihood of misclassification, thereby improving accuracy without sacrificing overall system speed.

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If the system lowers probability values for unavailable articles to reduce misclassification, then the accuracy for available articles is maintained, but the ability to detect actually-present unavailable articles deteriorates

Engineering Contradiction:
Improvearticle type inference accuracyVSAvoidsystem adaptability to actual article presence
Core Design Contradiction:
Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The system applies partial suppression to probability values for unavailable articles rather than complete exclusion. By lowering (but not eliminating) the probability values, the system maintains the possibility of detecting actually-present unavailable articles while still reducing the overall risk of misclassification. This partial action preserves adaptability while improving accuracy.

Inventive Principle:
Principle #16Partial or excessive action

Solution Approach 2:

The system incorporates feedback mechanisms where user corrections and verification results are used to continuously refine the probability adjustment strategy. When users confirm or correct article identifications, the system learns from these feedback signals to optimize the balance between reducing misclassification and maintaining detection capability for edge cases.

Inventive Principle:
Principle #23Feedback

Data Source

PatentEP3751456B1Article discrimination system and checkout processing system including article discrimination system
Publication Date: 2026.03.04 ISHIDA CO LTD
  • EP3751456B1 patent drawingFigure 1
  • EP3751456B1 patent drawingFigure 2
  • EP3751456B1 patent drawingFigure 3

AI summary

[Object] To provide an article discrimination system that can accurately infer the type of an article from an article image. [Solution] An article discrimination system includes an imager, an inference component, and a setting component. The imager captures an image of an article to acquire an article image. The inference component acquires first information which the inference component utilizes to infer the type of the article from the article image and, based on the first information acquired, infers one or plural types for the type of the article from among an article type group. The setting component sets at least one of types of articles that are available and types of articles that are not available among the article type group. The inference component preferentially infers, as the type of the article corresponding to the article image, the types of articles that are available over the types of articles that are not available.