Inspection Device Domain Adaptation Without Model Relearning

Solution Overview

Problem

Existing inspection devices face high relearning costs and production downtime due to differences in capturing conditions between the learning model creation site and the actual inspection site, leading to inaccurate output results when input images significantly deviate from the training data distribution.

Innovation Solution

An inspection device equipped with a camera, image inference means, and domain adaptation means that adapt image data to the distribution of learning data using metadata about capturing conditions, allowing for continued high-accuracy inspections without full relearning.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If relearning is performed to maintain high accuracy when distribution changes, then inspection accuracy is maintained, but relearning cost and production downtime increase

Engineering Contradiction:

Improveinspection accuracyVSAvoidproduction downtime

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The patent performs domain adaptation in advance by preparing adapted image data that bridges the distribution gap between training data and actual inspection data. This preliminary adaptation ensures the learning model can handle distribution shifts without requiring relearning, thus maintaining inspection accuracy while avoiding production downtime.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent introduces adapted image data as an intermediary between the original training data and the actual inspection data. This intermediate representation has a distribution that is neither exactly the training distribution nor the target distribution, but a blended distribution that allows the model to generalize better without relearning.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Measurement precision

If relearning is performed to maintain high accuracy when distribution changes, then inspection accuracy is maintained, but relearning cost increases

Engineering Contradiction:

Improveinspection accuracyVSAvoidrelearning cost

Core Design Contradiction:

Measurement precisionVSEase of manufacture

Solution Approach 1:

The patent performs domain adaptation in advance by preparing adapted image data that bridges the distribution gap between training data and actual inspection data. This preliminary adaptation ensures the learning model can handle distribution shifts without requiring relearning, thus maintaining inspection accuracy while avoiding production downtime.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent introduces adapted image data as an intermediary between the original training data and the actual inspection data. This intermediate representation has a distribution that is neither exactly the training distribution nor the target distribution, but a blended distribution that allows the model to generalize better without relearning.

Inventive Principle:

Principle #24Intermediary (Mediator)

3Productivity

If inspection is performed with significantly changed distribution, then continuous operation is maintained, but output result accuracy decreases

Engineering Contradiction:

Improvecontinuous operationVSAvoidoutput result accuracy

Core Design Contradiction:

ProductivityVSMeasurement precision

Solution Approach 1:

The patent performs domain adaptation in advance by preparing adapted image data that bridges the distribution gap between training data and actual inspection data. This preliminary adaptation ensures the learning model can handle distribution shifts without requiring relearning, thus maintaining inspection accuracy while avoiding production downtime.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent changes the distribution parameters of the image data by creating adapted image data that combines features from both training and target domains. This parameter transformation allows the model to maintain accuracy when processing data with significantly changed distribution characteristics.

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS20250265696A1Inspection device and inspection method

Publication Date: 2025.08.21 ANRITSU CORP

AI summary

There is provided an inspection device including: a camera that captures an image of an inspection object to acquire image data; image inference means for outputting a determination result indicating a quality of the inspection object based on the input image data; and domain adaptation means for outputting adapted image data, which is adapted to a distribution of learning image data used for model generation by the image inference means, based on first image data obtained by capturing the image of the inspection object and metadata indicating a distribution of the first image data.