Feature Selection Using Expert and Non-Expert Defect Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods, such as random forest, struggle to accurately select feature quantities with high contribution degrees in classification models for defect detection when sufficient teacher data is not available, particularly in the initial stages of production lines, leading to inefficiencies and inaccurate defect classification.
Innovation Solution
A method and device for selecting feature quantities with high contribution degrees using a combination of expert and non-expert data, involving extraction of multidimensional features, calculation of standard deviations, and determination of standard deviation ratios to identify key features for specific defect modes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If random forest is used to select feature quantities with high contribution degrees, then feature selection can be performed, but accurate selection is not achieved when sufficient teacher data is not available
Solution Approach 1:
The patent applies preliminary action by pre-calculating and storing contribution degree information for multiple candidate feature quantities across different defect modes before actual defect detection. This allows the system to quickly retrieve and combine pre-computed contribution degrees from expert and non-expert data, enabling accurate feature selection even when current teacher data is insufficient.
Solution Approach 2:
The patent introduces contribution degree as an intermediary metric that bridges expert data and non-expert data. By calculating contribution degrees separately for both data types and combining them, the system uses this intermediary measure to identify feature quantities that are consistently important across different data sources, achieving accurate feature selection with limited teacher data.
2Reliability
If only expert data is used to collect teacher data, then data quality is high, but collection cost increases due to requiring skilled persons for long hours
Solution Approach 1:
The patent merges expert data and non-expert data by calculating and combining their respective contribution degrees. This combination allows the system to leverage the high quality of expert data while incorporating the abundant quantity of non-expert data, achieving both data reliability and collection efficiency simultaneously.
Solution Approach 2:
The patent changes the evaluation parameter from direct data quality assessment to contribution degree calculation. By evaluating feature quantities based on their contribution degrees derived from both expert and non-expert data, the system transforms the problem of balancing data quality and collection cost into a mathematical optimization problem that can be solved objectively.
3Productivity
If only non-expert data is used to collect teacher data, then collection cost is low, but classification accuracy decreases for low occurrence probability defects
Solution Approach 1:
The patent uses contribution degree as an intermediary to filter and select feature quantities from non-expert data. By identifying feature quantities with high contribution degrees in both expert and non-expert data, the system ensures that even rare defects with low occurrence probability are captured through features that consistently show high importance across different data sources.
Solution Approach 2:
The patent implements feedback by using the contribution degree calculation results to guide feature quantity selection. The system feeds back the identified high-contribution feature quantities into the defect detection model, ensuring that the model focuses on the most discriminative features for accurate classification of rare defects.
4Measurement precision
If sufficient teacher data is prepared beforehand for random forest, then accurate feature quantity selection is achieved, but data collection time and cost increase
Solution Approach 1:
The patent applies preliminary action by pre-calculating contribution degrees for multiple candidate feature quantities across different defect modes and storing them for quick retrieval. This allows the system to perform accurate feature quantity selection without requiring extensive data collection and processing at the time of detection, significantly reducing time loss.
Data Source
AI summary
A feature quantity selection method according to the present invention includes extracting a multidimensional feature quantity from expert data including images of various defect shapes, extracting a multidimensional feature quantity from non-expert data including images of limited defect shapes, calculating a standard deviation in every dimension of the respective feature quantities of the expert data and the non-expert data that have been extracted; calculating a ratio between the standard deviation in every dimension of the calculated expert data and the standard deviation in every dimension of the calculated non-expert data, selecting a predetermined number of standard deviation ratios in descending order of value from among the standard deviation ratios that have been calculated, and selecting feature quantities associated with the standard deviation ratios that have been selected, as the feature quantities each having a high contribution degree to a specific defect mode.