Defect Classification Apparatus with Weight Coefficient Adjustment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing defect classification technologies, even with hierarchical classification models, face reliability issues when purity is below a certain threshold, requiring manual verification and lacking clear identification of adjustments for improving classification performance.
Innovation Solution
A defect classification apparatus that evaluates and outputs classification performance using weight coefficients and likelihood values to identify areas for adjustment, allowing for the refinement of classification models and improving reliability by distinguishing sub-classes with low purity and high misclassification rates.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If hierarchical classification model is used, then classification performance is improved, but reliability is insufficient when purity is below threshold
Solution Approach 1:
The patent introduces a feedback mechanism where classification results are evaluated against reference data, and weight coefficients are adjusted based on evaluation outcomes. When purity is below threshold, the system automatically adjusts weight coefficients to improve reliability, creating a closed-loop feedback system that continuously enhances classification performance.
Solution Approach 2:
The patent dynamically changes the weight coefficients of classification parameters based on evaluation results. By adjusting these parameters according to measured purity levels and classification performance, the system adapts to maintain reliability even when initial classification accuracy is insufficient.
2Productivity
If automatic classification is performed without evaluation output, then productivity is improved, but reliability cannot be ensured
Solution Approach 1:
The system incorporates an evaluation output mechanism that provides feedback on classification reliability. This allows automatic classification to proceed efficiently while simultaneously monitoring and reporting reliability metrics, enabling users to trust or verify results as needed without manual intervention for every case.
Solution Approach 2:
The classification system performs self-evaluation and self-adjustment by automatically computing purity metrics and adjusting weight coefficients without requiring external manual verification. This self-service capability maintains high productivity while ensuring reliability through automated quality control.
3Reliability
If weight coefficients are manually adjusted, then classification reliability is improved, but operation complexity increases
Solution Approach 1:
The system automatically adjusts weight coefficients based on evaluation results without requiring manual intervention. The classification apparatus performs self-tuning by computing optimal weights from reference data and evaluation outcomes, eliminating complex manual adjustment operations while maintaining high reliability.
Solution Approach 2:
The automated weight coefficient adjustment is driven by feedback from classification evaluation. The system continuously monitors performance and automatically modifies weights in response to evaluation results, replacing manual adjustment operations with an autonomous feedback-controlled process that simplifies operation while improving reliability.
Data Source
AI summary
Disclosed is a technique wherein an object that requires adjustment in order to increase the reliability of automatic classification can be easily identified. A device (140) for adjusting classification classifies defects into a first class group according to the feature amount of the defects that are obtained from image data obtained from an electron microscope (110), and classifies the defects into a second class group according to the feature amount of the defects classified into the first class group. And, the device (140) for adjusting the classification calculates classification performance by comparing the defects that have been classified into the second class group, and outputs the calculated classification performance in a predetermined display format to an output unit (180).