AI Defect Inspection Using Material-Aware Image Reconstruction

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Image generation AI systems face challenges in accurately reconstructing normal images, leading to false positive determinations of defects in inspection images, particularly when the inspection region overlaps with material boundaries or includes base layers that are difficult to reconstruct.

Innovation Solution

A determination apparatus and method that incorporates material information into the mask as color information and removes base layer regions during error calculation to improve the reconstruction process, using trained image generation AI to accurately determine defects.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If image generation AI is used to reconstruct normal images for defect detection, then the system can detect new types of defects, but false positive determinations increase when material boundaries or base layers are present in the inspection region

Engineering Contradiction:

Improvedefect detection capabilityVSAvoiddetermination accuracy

Core Design Contradiction:

Adaptability or versatilityVSReliability

Solution Approach 1:

The inspection region is segmented by material types using color-coded masks. Each material region is processed separately during reconstruction, allowing the AI to handle different material characteristics appropriately and reduce false positives at material boundaries.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

Different reconstruction strategies are applied to different material regions. The mask coloring provides local information about material types, enabling the AI to adjust reconstruction quality and parameters according to the specific material being inspected.

Inventive Principle:

Principle #3Local quality

2Productivity

If a simple mask is used for reconstruction, then the processing is faster and simpler, but the reconstruction accuracy decreases leading to more false positive determinations

Engineering Contradiction:

Improveprocessing efficiencyVSAvoidreconstruction accuracy

Core Design Contradiction:

ProductivityVSMeasurement precision

Solution Approach 1:

The mask uses color information to encode material type data. This allows rich information to be conveyed in a compact visual format that the AI can process efficiently during reconstruction, improving accuracy without significantly increasing computational burden.

Inventive Principle:

Principle #32Color changes

Solution Approach 2:

Material information is pre-encoded into the mask before the reconstruction process. This preliminary preparation of information allows the AI to perform more accurate reconstruction without requiring additional processing steps during the main inspection workflow.

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS20260017777A1Determination apparatus, training apparatus, determination method, training method, determination program, and training program

Publication Date: 2026.01.15 NITTO DENKO CORP

AI summary

A determination apparatus includes a processor configured to: use a trained image generation AI-trained so as to reconstruct first image from first mask image in which mask is overlaid onto inspection region of the first image, the mask being colored according to types of material included in corresponding region of inspection target object and being configured to be overlaid onto the inspection region, the first image being image determined not to contain defect among captured images of the inspection target object; and compare second reconstruction image with second image to determine whether or not the second image contains defect, the second reconstruction image being reconstructed by inputting second mask image into the trained image generation AI, the second mask image being image in which the mask is overlaid on, and corresponds to, inspection region of the second image, the second image being captured image of the inspection target object.