Synthetic Defect Image Generation for Inspection Coverage
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing defect inspection models fail to achieve complete coverage of defect feature combinations, leading to inadequate identification and characterization of defects, especially with limited training data.
Innovation Solution
A computer-implemented method and system that trains a model using defect images with encoded textual features to generate synthetic images representing distinct feature combinations, enhancing coverage and enabling the detection of new defect types through a generative adversarial network (GAN) model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a model is trained using a limited number of defect feature combinations, then the training process is simpler and faster, but the model cannot achieve complete coverage of all defect types
Solution Approach 1:
The system performs preliminary action by generating synthetic defect images for all possible feature combinations before actual defect inspection. This pre-generation of comprehensive training data ensures that the model is exposed to all potential defect types upfront, eliminating the need for extensive real-world data collection and enabling complete coverage without compromising training efficiency
Solution Approach 2:
The system creates synthetic copies of defect images by combining features from existing real defect images. Instead of requiring unique real samples for every possible feature combination, the system generates realistic synthetic representations by copying and recombining visual features, thereby achieving complete coverage while maintaining efficient training
2Adaptability or versatility
If more defect images with diverse feature combinations are used for training, then complete coverage of defect types is achieved, but the data collection and processing becomes more complex
Solution Approach 1:
The system uses copying to generate synthetic defect images by recombining visual features from existing real defect images. This approach creates diverse feature combinations without requiring complex data collection processes, as the synthetic images are generated algorithmically from available data
Solution Approach 2:
The system changes parameters by systematically varying feature combinations in synthetic image generation. By controlling which features are combined and how, the system generates diverse defect types with precise parameter control, avoiding the complexity of collecting and managing large volumes of real-world diverse defect data
Data Source
AI summary
Aspects described herein include a computer-implemented method and associated system and computer program product. The method includes training a model using a plurality of defect images. Each defect image corresponds to a respective first feature combination of encoded textual features of a predefined set of textual features. The method further includes generating a first synthetic image using the model. The first synthetic image corresponds to a second feature combination of encoded textual features of the predefined set that is distinct from the first feature combinations.