Image Determination Training With Accuracy-Based Data Selection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing image determination systems face increased time and computational demands when additional training is performed to improve accuracy, despite minimal improvement in determination accuracy.

Innovation Solution

The system divides training data into sub-training data, trains using each subset, and selects the most accurate subset for further training, reducing the number of data points required while maintaining or improving determination accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If additional training data is used to train the model, then determination accuracy may be improved, but training time and computational load increase

Engineering Contradiction:
Improvedetermination accuracyVSAvoidtraining time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent segments the training data into multiple subsets and trains multiple models on different subsets. This allows the system to evaluate multiple training approaches in parallel and select the best performing model without requiring all data to be processed sequentially, thereby reducing overall training time while maintaining accuracy.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system automatically evaluates multiple training subsets and selects the optimal model without requiring manual intervention. The determination device autonomously compares accuracy metrics across different trained models and chooses the best performing one, reducing the need for human expertise in model selection while maintaining high accuracy.

Inventive Principle:
Principle #25Self-service

2Measurement precision

If additional training data is used to train the model, then determination accuracy may be improved, but computational load increases

Engineering Contradiction:
Improvedetermination accuracyVSAvoidcomputational load
Core Design Contradiction:
Measurement precisionVSUse of energy by moving object

Solution Approach 1:

By dividing training data into multiple subsets and training separate models on each subset, the computational load for each individual training process is reduced. The system can then select the best model from these lighter computational tasks rather than processing all data in a single heavy training operation.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent extracts and selects only the most effective training subset for final model deployment. Instead of using all training data continuously, the system identifies and extracts the optimal subset that provides the best accuracy, thereby reducing the computational load required for maintaining the model.

Inventive Principle:
Principle #2Taking out (Extraction)

3Reliability

If more training data is used, then model robustness may improve, but training complexity increases

Engineering Contradiction:
Improvemodel robustnessVSAvoidtraining complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The training process is segmented into multiple independent subset trainings, each with its own model. This modular approach simplifies the management of large datasets by breaking them into manageable pieces, reducing training complexity while still achieving robustness through ensemble selection.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system automatically manages the complexity of multiple training subsets through self-evaluation and automatic selection. The determination device independently assesses each trained model's accuracy and autonomously selects the best one, eliminating the need for complex manual management of training configurations.

Inventive Principle:
Principle #25Self-service

Data Source

PatentEP3905191B1Image determination device, method and program
Publication Date: 2025.10.15 OMRON CORP
  • EP3905191B1 patent drawingFigure 1
  • EP3905191B1 patent drawingFigure 2
  • EP3905191B1 patent drawingFigure 3

AI summary

Provided is an image determination device or the like which can suppress an increase in time or a calculation amount necessary for a training process, even when a training model is further trained. The image determination device is provided with: a training model which outputs, on the basis of an image to be examined, output data indicating a determination result about the image; a training unit which trains the training model to output, by using training data including a training image and label data, output data indicating the label data associated with the training image, when the training image is input to the training model; a dividing unit which divides the training data into a plurality of pieces of sub-training data; a measurement unit which measures accuracy of determination when the training unit trains the training model by using each of the plurality of pieces of sub-training data; and selection unit which selects at least any one among the plurality of pieces of sub-training data on the basis of the accuracy of determination.