Defect Detector Training Using Reconstruction Model
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing methods for training defect detectors in computer surface inspection require a large amount of labeled training data, which is costly and impractical due to labor-intensive data collection, security concerns, and the need for diverse and consistent labeling, especially as product designs become more diverse and lifecycles shorten.
Innovation Solution
A method that uses a small amount of human-labeled image data by obtaining reference images with defects, training a reconstruction model for tolerable defect levels, generating labels for target images through error calculation procedures, and training a defect detector using machine learning algorithms, thereby reducing the need for extensive labeled data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a large amount of labeled training data is collected to improve defect detection accuracy, then the detection precision is improved, but the labor cost and time consumption increase significantly
Solution Approach 1:
The patent uses a reconstruction model to generate synthetic defect images by copying and transforming normal surface patterns. These synthesized images serve as training data, eliminating the need to manually collect and label large amounts of real defect images. The reconstruction model creates realistic defect variations through digital manipulation of normal surface data.
Solution Approach 2:
The system performs self-labeling by using the reconstruction model to automatically generate both the defect images and their corresponding labels. The error calculating procedure automatically identifies defect regions and generates annotation data without human intervention, making the data preparation process self-sufficient.
2Adaptability or versatility
If diverse training data is collected to cover various product designs and defect types, then the adaptability of the detector is improved, but the complexity of data management and labeling consistency increases
Solution Approach 1:
The reconstruction model serves multiple functions: it generates defect images for various product types, creates diverse defect variations, and produces consistent labels automatically. This single multi-functional system replaces the need for separate data collection and labeling processes for each product type and defect category.
Solution Approach 2:
The system achieves diversity by changing parameters within the reconstruction model, such as defect position, size, shape, and intensity. By manipulating these parameters, the model generates varied defect scenarios from a single normal surface image, covering multiple product designs and defect types without external data collection.
3Reliability
If manual data collection and labeling are performed to ensure data quality, then the reliability of training data is improved, but the production efficiency decreases
Solution Approach 1:
The patent replaces the manual mechanical process of data collection and labeling with an automated computational system. The reconstruction model and error calculating procedure form an automated pipeline that generates and annotates training data without human physical intervention, dramatically improving productivity while maintaining reliability through consistent algorithmic processing.
Data Source
AI summary
A method for training a defect detector comprises: obtaining a first reference image of a first reference object, wherein the first reference object has a defect and the first reference image has a first label indicating the defect; training a reconstruction model according to a second reference image of a second reference object associated with the first reference object, wherein a defect level of the second reference object is in a tolerable range with an upper limit; obtaining a target image of a target object associated with the first reference object and the second reference object; generating a second label according to the target image, the reconstruction model and an error calculation procedure, wherein the second label comprises a defect of the target object; and training a defect detector by performing a machine learning algorithm according to the first reference image, the target image and the second label.