Article Recognition Using Shared Feature Vectors for Inventory Changes

Solution Overview

Problem

Conventional self-checkout systems face challenges in accurately identifying a wide variety of products due to frequent changes in inventory, requiring extensive data re-labeling and retraining of neural networks, which is impractical and decreases recognition accuracy.

Innovation Solution

An article recognition system and method where the article detection, feature extraction, and discrepancy determination processes are performed separately, using a shared feature extraction model across phases, with minimal data preparation, including image registration and discrepancy determination to compare feature vectors.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional neural network retraining is performed for every inventory change, then recognition accuracy can be maintained, but the system complexity and time consumption increase significantly

Engineering Contradiction:

Improverecognition accuracyVSAvoidsystem complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent divides the article recognition process into three independent phases: article detection phase (detecting articles in images), shared feature extraction phase (extracting features using a pre-trained model), and discrepancy determination phase (comparing features to identify articles). This segmentation allows the feature extraction model to be trained once and reused across multiple inventory changes, eliminating the need for complete retraining while maintaining recognition accuracy.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent performs feature extraction model training in advance during a preparation phase, before actual article recognition is needed. The pre-trained model is then reused in subsequent article detection and feature extraction phases, allowing the system to adapt to inventory changes without retraining, thus reducing system complexity and time consumption.

Inventive Principle:

Principle #10Preliminary action

2Measurement precision

If extensive data re-labeling is performed for inventory changes, then recognition accuracy can be maintained, but the time and labor resources required increase

Engineering Contradiction:

Improverecognition accuracyVSAvoidtime consumption

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The patent performs feature extraction model training in advance during a preparation phase, before actual article recognition is needed. The pre-trained model is then reused in subsequent article detection and feature extraction phases, allowing the system to adapt to inventory changes without retraining, thus reducing system complexity and time consumption.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent creates a pre-trained feature extraction model that can be copied and reused across different inventory scenarios. Instead of creating new models for each inventory change, the same pre-trained model is applied to extract features from articles in different phases, significantly reducing the time and labor required for data preparation.

Inventive Principle:

Principle #26Copying

3Adaptability or versatility

If the same feature extraction model is used across phases, then system adaptability improves, but the initial model training time increases

Engineering Contradiction:

Improvesystem adaptabilityVSAvoidtraining time

Core Design Contradiction:

Adaptability or versatilityVSLoss of time

Solution Approach 1:

The patent performs feature extraction model training in advance during a preparation phase, before actual article recognition is needed. The pre-trained model is then reused in subsequent article detection and feature extraction phases, allowing the system to adapt to inventory changes without retraining, thus reducing system complexity and time consumption.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent divides the article recognition process into three independent phases: article detection phase (detecting articles in images), shared feature extraction phase (extracting features using a pre-trained model), and discrepancy determination phase (comparing features to identify articles). This segmentation allows the feature extraction model to be trained once and reused across multiple inventory changes, eliminating the need for complete retraining while maintaining recognition accuracy.

Inventive Principle:

Principle #1Segmentation

Data Source

PatentUS20260065636A1Article Recognition System and Method

Publication Date: 2026.03.05 FLYTECH TECH

AI summary

The present invention relates to an article recognition method. The method includes of executing a first feature extraction programming module to extract a plurality of candidate article feature vectors from a plurality of candidate article images; executing a second feature extraction programming module and an image registration transformation programming unit to perform an image registration on a target article image and extract a target article feature vector therefrom; and executing a discrepancy determination programming module to compare the target article feature vector with each of the plurality of candidate article feature vectors and generate a similarity score accordingly.