Defect Teacher Data Generation with Self-Annotating Boundary Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for generating teacher data for inspection devices with machine learning functions, particularly for defective products, are costly and prone to accuracy issues due to the scarcity of defective product data and the need for extensive annotation.
Innovation Solution
A teacher data generating method that uses a generation model to create annotated defective product data by superimposing boundary information indicating the defect range in chromatic color onto gray-scale defective product images, allowing for the extraction and calculation of accurate defect coordinates without requiring a classification model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a classification model is used to annotate defective product data, then the annotation accuracy depends on the model accuracy, but the cost increases and the process becomes complex
Solution Approach 1:
The patent extracts the boundary information (annotation) directly from the input defective product images during the generation process, rather than using a separate classification model to annotate generated images. This eliminates the need for complex model construction while maintaining annotation accuracy.
Solution Approach 2:
The generation model automatically provides its own annotation through the boundary information embedded in the generated images. The system self-annotates without requiring external classification models, reducing complexity while maintaining accuracy.
2Reliability
If many pieces of defective product data are collected for learning, then the inspection accuracy improves, but the data collection becomes difficult due to scarcity of defective products
Solution Approach 1:
The patent performs preliminary generation of defective product data using the generation model trained on limited real defective data. This preliminary action creates synthetic defective data that can be used for training inspection models, overcoming the scarcity of real defective products.
Solution Approach 2:
The generation model creates copies of defective product patterns from limited real examples, generating synthetic defective data that replicates the characteristics of real defects. This allows abundant training data to be created from scarce real defective samples.
3Manufacturing precision
If annotation operation is performed for enormous amount of defective product data, then the learning quality improves, but the time and cost required increases significantly
Solution Approach 1:
The generation model automatically generates boundary information (annotations) as part of the image generation process itself. This self-service approach eliminates the need for separate manual or model-based annotation operations, dramatically reducing time and cost while maintaining learning quality.
Solution Approach 2:
The patent merges the image generation process with the annotation generation process into a single unified operation. The boundary information is generated simultaneously with the defective product images, eliminating the need for separate annotation steps and reducing overall processing time.
Data Source
AI summary
A teacher data generating method includes: causing a generation model to perform learning by using a learning image in which boundary information indicating in a chromatic color a range of a defect is superimposed on a defective product image in a gray scale so as to generate a generated defect image including a new image of a defect in the gray scale and an image of the boundary information in the chromatic color; generating the generated defect image by using the generation model; extracting a pixel having a pixel value corresponding to the chromatic color from the generated defect image, extracting the boundary information corresponding to the generated defect image, and acquiring a gray scale defect image without including an image of the boundary information; calculating a coordinate of the boundary information; and associating the gray scale defect image with the coordinate to obtain defective product teacher data.