Automated Teacher Data Generation for Product Inspection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing defect classification methods require significant human labor and are prone to errors due to manual category assignment, leading to increased costs and decreased accuracy in generating teacher data for machine learning models used in inspecting product quality.
Innovation Solution
A learning device that utilizes a hyperspectral camera to acquire physical property information, automatically assigning categories to image data, thereby reducing human intervention and generating a large amount of teacher data for creating a machine learning model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual category assignment is used to generate teacher data, then human labor and time are required, but this increases costs and decreases accuracy due to human errors
Solution Approach 1:
The patent replaces the manual mechanical process of category assignment with an automated system using physical property measurement and machine learning. The processing unit automatically determines categories by comparing measured physical properties against reference data and learned models, eliminating human manual assignment while improving accuracy and reducing time consumption.
Solution Approach 2:
The system enables self-service by allowing the learning model to automatically generate teacher data without human intervention. The processing unit uses the learned model to autonomously categorize image data, creating teacher data that can be continuously generated and updated without requiring manual human effort.
2Productivity
If manual category assignment is used, then human expertise can be applied, but the process is prone to errors and cannot scale efficiently
Solution Approach 1:
The patent replaces manual category assignment with an automated processing system that uses physical property measurement and machine learning models. This substitution eliminates human errors and inconsistencies while enabling scalable teacher data generation, improving both productivity and reliability simultaneously.
Solution Approach 2:
The system implements feedback mechanisms where the learning model is continuously improved using generated teacher data. The processing unit refines the learning model based on the accuracy of category assignments, creating a self-improving system that increases reliability over time while maintaining high productivity.
3Quantity of substance
If a large amount of teacher data is generated manually, then more training data is available, but the burden of creation increases significantly
Solution Approach 1:
The system enables self-service by automatically generating teacher data through the processing unit's categorization process. The learning model autonomously creates labeled training data without human intervention, allowing large quantities of teacher data to be generated easily and continuously, significantly reducing the burden of data creation.
Solution Approach 2:
The system performs preliminary action by pre-processing image data through physical property measurement and automatic categorization before the actual machine learning training. This preliminary automated processing prepares the teacher data in advance, making the overall process easier and more efficient while enabling large-scale data generation.
Data Source
AI summary
A learning device includes a camera configured to acquire image data by imaging a sample of a product, a physical property information acquisition unit configured to acquire physical property information of the sample, and a processing unit configured to generate a learning model. The processing unit is configured to identify a category of the sample based on rule information relating the physical property information to the category, to generate teacher data by relating the identified category to the image data, and to generate a learning model by machine learning using the teacher data. The learning model outputs the category of the sample in response to an input of the image data of the sample.