Display Panel Inspection AI Training With Synthetic Defect Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for inspecting display panels using artificial intelligence models are limited by the need for extensive collection of bad quality images, which is time-consuming and prone to errors, and lack effective training strategies to differentiate between good and bad quality images.
Innovation Solution
A learning method that trains a first artificial intelligence model on good quality images and a second model on a combination of good and synthetic bad quality images, using feature vector extraction and layer modifications to enhance differentiation and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If real bad quality images are collected for training, then the model can learn actual defect patterns, but the process is time-consuming and prone to errors
Solution Approach 1:
The patent uses synthetic bad quality images generated by combining defect patterns with good quality images instead of collecting real defective images. This copying approach creates training data that mimics real defects without the time-consuming collection process, resolving the contradiction between detection accuracy and data collection time
Solution Approach 2:
The patent pre-generates defect patterns and prepares synthetic bad quality images before model training. By performing the data preparation action in advance through automated synthesis rather than manual collection, it eliminates the time-consuming aspect while maintaining the quality needed for accurate defect detection
2Device complexity
If a single AI model is trained on both good and bad quality images, then the training process is simpler, but the model struggles to differentiate between good and bad quality images
Solution Approach 1:
The patent divides the training process into two separate models: a first model trained only on good quality images to learn normal patterns, and a second model trained on both good and bad quality images to learn differentiation. This segmentation allows each model to specialize, improving the second model's ability to differentiate while keeping the overall system manageable
Solution Approach 2:
The first AI model serves as an intermediary that extracts feature vectors from images, which are then used as input for the second model. This intermediary processing step helps the second model focus on learning the differences between good and bad quality images by providing pre-processed features, thereby improving differentiation capability
3Measurement precision
If the second model includes all layers of the first model, then feature extraction is preserved, but the model size and training complexity increase
Solution Approach 1:
The second AI model includes only the necessary layers from the first model that are required for feature extraction, rather than copying all layers. This selective extraction maintains the essential feature extraction capability while reducing the overall model size and training complexity of the second model
Data Source
AI summary
In a learning method of an artificial intelligence model for inspecting a display panel, the learning method includes training a first artificial intelligence model using good quality images of the display panel, and training a second artificial intelligence model including at least a partial layer of the first artificial intelligence model to learn, using bad quality images, different than the good quality images. At least one of the bad quality images is a synthetic image in which a target portion is combined with a base image.