Stepwise Discriminator Training for Product Inspection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional machine learning techniques for determining product acceptability using discriminators like neural networks often fall into local solutions, leading to low generalization ability and precision, especially when biased towards specific defects or difficult to distinguish cases, even with sufficient training data.
Innovation Solution
An inspection system that acquires learning data sets with varying difficulty levels by utilizing multiple discriminators to set difficulty levels based on their output conformity to correct answers, and constructs a second discriminator through stepwise machine learning, ascending the difficulty level of learning data sets to avoid local solutions and enhance generalization performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If machine learning is performed using a great amount of learning data at once, then the training completeness is improved, but the discriminator falls into local solutions and generalization ability deteriorates
Solution Approach 1:
The patent segments the learning data into multiple groups based on difficulty levels (easy, medium, hard) and trains discriminators in sequential stages rather than using all data at once. This segmentation prevents the discriminator from falling into local solutions while still utilizing the complete dataset for comprehensive training.
Solution Approach 2:
The patent performs preliminary classification of learning data into difficulty-based groups before training. By preparing and organizing the data in advance according to difficulty levels, the system enables staged training that progressively builds discriminator capability without overwhelming it with all data simultaneously.
2Productivity
If learning data is biased towards a specific defect type, then training efficiency for that defect is improved, but the discriminator cannot determine other defect types
Solution Approach 1:
The patent applies local quality by creating specialized training groups for different defect types (e.g., dents, scratches, dirt) with varying difficulty levels. Each group focuses on specific defect characteristics while maintaining appropriate difficulty progression, allowing the discriminator to develop specialized expertise in different defect categories.
Solution Approach 2:
The patent creates a universal training framework that handles multiple defect types through a common difficulty-based segmentation approach. The same staged training methodology is applied across different defect categories, enabling the discriminator to generalize across various defect types while maintaining high efficiency for each specific type.
3Measurement precision
If only learning data with difficult discrimination cases is collected, then the discriminator's ability to handle hard cases is improved, but the machine learning process takes excessive time
Solution Approach 1:
The patent implements periodic action through staged training cycles where the discriminator is trained on easy cases first to establish baseline performance, then progressively exposed to medium and hard cases. This periodic progression through difficulty levels optimizes learning time by building foundational skills before tackling complex discrimination tasks.
Solution Approach 2:
The patent performs preliminary organization of learning data into difficulty-based groups before training begins. By pre-classifying data and preparing staged training groups in advance, the system eliminates time-wasting random sampling and ensures efficient progression through increasingly difficult cases.
4Measurement precision
If the discriminator determines small stains as defective products, then detection sensitivity is improved, but false positive rate increases
Solution Approach 1:
The patent changes the training parameter of difficulty level progression to teach the discriminator appropriate sensitivity thresholds. By exposing the discriminator to staged examples including subtle defects like small stains with increasing difficulty, the system fine-tunes the discriminator's sensitivity to distinguish between acceptable and defective small features, reducing false positives while maintaining detection capability.
Data Source
AI summary
A characteristic included in data is determined with relatively high precision. An inspection system according to one aspect of the present invention acquires a plurality of learning data sets respectively including a combination of image data and correct answer data, and sets a difficulty level of determination for each of the learning data sets in accordance with a degree to which a result which is obtained by determining the acceptability of a product in the image data of each of the learning data sets by a first discriminator conforms to a correct answer indicated by the correct answer data. Besides, the inspection system constructs a second discriminator that determines the acceptability of the product by executing stepwise machine learning in which the learning data sets are utilized in ascending order of the set difficulty level.