Neural Network Artificial Defect Data Generation for Vision Inspection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In automated vision-based control systems for industrial products, the scarcity of defect data leads to an unbalanced dataset, affecting the performance of defect recognition classification models. Existing methods for oversampling or undersampling can lose original data information and lead to overfitting.
Innovation Solution
A method using a neural network model to generate artificial defect data, which involves obtaining input data, dividing it into sub-data, generating artificial defect data using a first neural network model, transforming the sub-data into the artificial defect data, and obtaining final artificial defect data through smoothing with a second neural network model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If oversampling or undersampling is performed on defect data, then data imbalance is addressed, but original data information is lost and overfitting occurs
Solution Approach 1:
The patent uses a neural network model to generate artificial defect data that copies the characteristics of real defect data. The generator network creates synthetic defect images that preserve original data information while increasing dataset quantity, avoiding the information loss associated with traditional sampling methods.
Solution Approach 2:
The patent transforms defect data through parameter changes by using a neural network to generate new defect images with controlled characteristics. The generator network modifies data parameters to create artificial defects that maintain information integrity while balancing the dataset.
2Quantity of substance
If traditional sampling methods are used to balance datasets, then data quantity is adjusted, but data diversity decreases and overfitting occurs
Solution Approach 1:
The neural network generator creates artificial defect data that copies the diversity and characteristics of real defect data. This ensures that the generated data maintains high diversity while increasing the overall dataset quantity, preventing overfitting.
Solution Approach 2:
The patent divides the defect data into different categories and uses the neural network to generate artificial data for each category separately. This segmentation approach maintains data diversity across different defect types while balancing the overall dataset.
3Quantity of substance
If defect data is augmented using traditional methods, then dataset size increases, but original pixel values are lost and model performance deteriorates
Solution Approach 1:
The neural network model generates artificial defect data that preserves the quality and pixel value information of original defect data. This copying approach maintains measurement precision for defect recognition while increasing dataset size.
Solution Approach 2:
The patent uses parameter changes in the neural network to generate artificial defect data that maintains the essential characteristics and pixel values of original data. This ensures that defect recognition accuracy is preserved while expanding the dataset.
Data Source
AI summary
Disclosed is a method for generating artificial defect data by using a neural network model, which is performed by one or more processors of a computing device according to an exemplary embodiment of the present disclosure.The method may include: obtaining input data, and dividing the input data into a plurality of sub-data; generating one or more artificial defect data by using a first neural network model; transforming the one or more sub-data into the generated artificial defect data; and obtaining final artificial defect data based on a plurality of sub-data including the one or more transformed sub-data.