Inspection Apparatus for Defect Detection Using Targeted Learning Data Generation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing image identification devices face challenges in achieving high identification accuracy for defects in objects, as the quality and variety of learning data significantly impact their performance, and it is unclear which types of images should be added to improve accuracy.

Innovation Solution

A data generation apparatus and method that generates new learning data by analyzing incorrect determinations made by the identification device, focusing on partial images that significantly influence the decision, and creating composite images to increase data variation, thereby reducing erroneous determinations and improving identification accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If learning data includes sufficient number and types of images for high identification accuracy, then identification accuracy improves, but data generation complexity and time increase

Engineering Contradiction:
Improveidentification accuracyVSAvoiddata generation time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The system performs preliminary analysis of incorrect determinations to identify problematic partial images before generating new learning data. By pre-identifying which partial images contribute most to erroneous determinations, the system can target data generation efficiently without exhaustive analysis of all images, thus improving identification accuracy while controlling data generation time.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

Instead of uniformly processing all images, the system focuses on specific local regions (partial images) that have the greatest impact on incorrect determinations. By extracting and analyzing only the relevant partial images that contribute significantly to errors, the system efficiently generates targeted learning data that improves identification accuracy without requiring extensive processing of entire image sets.

Inventive Principle:
Principle #3Local quality

2Measurement precision

If comprehensive analysis of error causes is performed to generate targeted learning data, then identification accuracy improves, but computational complexity increases

Engineering Contradiction:
Improveidentification accuracyVSAvoidcomputational complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The system extracts only the essential elements needed for improving identification accuracy - specifically, partial images that significantly influence incorrect determinations. By separating and focusing on these critical partial images rather than analyzing entire images comprehensively, the system reduces computational complexity while still generating effective targeted learning data.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The system performs partial analysis by focusing on a subset of images that are most relevant to incorrect determinations, rather than comprehensively analyzing all images. By applying action to only the necessary portion (partial images with high influence on errors), the system achieves improved identification accuracy without the excessive computational burden of complete analysis.

Inventive Principle:
Principle #16Partial or excessive action

3Measurement precision

If diverse types of images are added to learning data to cover all defect scenarios, then identification accuracy improves, but data management complexity increases

Engineering Contradiction:
Improveidentification accuracyVSAvoiddata management complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The system applies local quality by generating diverse learning data focused on specific local regions (partial images) that are most relevant to incorrect determinations. Instead of managing comprehensive diverse image sets, the system creates targeted variations of problematic partial images, reducing data management complexity while improving identification accuracy for specific defect scenarios.

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The system performs preliminary identification of which partial images cause incorrect determinations before generating diverse learning data. This preliminary action allows the system to focus data generation on specific problematic regions, thereby reducing the overall complexity of data management while ensuring coverage of critical defect scenarios that need improvement.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP3480735B1Inspection apparatus, data generation apparatus, data generation method, and data generation program
Publication Date: 2023.07.19 OMRON CORP
  • EP3480735B1 patent drawingFigure 1
  • EP3480735B1 patent drawingFigure 2
  • EP3480735B1 patent drawingFigure 3

AI summary

Provided is an inspection apparatus, a data generation apparatus, and the like for generating learning data with which the identification accuracy of an identification device can be improved such that whether or not an object to be inspected includes a defect can be identified with high accuracy. An inspection apparatus includes: an image capturing apparatus configured to capture an image of an object to be inspected: a determination unit configured to determine, based on the image, whether or not the object to be inspected includes a defect, using an identification device that has been trained using learning data: an input unit configured to accept an input indicating whether or not a determination result by the determination unit is correct; an extraction unit configured to extract a partial image of the image based on which the determination has been made; and a generation unit configured to generate new learning data based on the partial image, if a fact that the determination result by the determination unit is not correct has been input.