Virtual Defective Image Synthesis for X-ray Inspection Absorptivity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image generation methods for X-ray inspection apparatuses fail to accurately reflect absorptivity for real electromagnetic waves, leading to inadequate machine learning performance when using virtual defective-product images synthesized from non-defective-product images of different energy bands.
Innovation Solution
An image generation device that synthesizes virtual defective-product images by adjusting pixel values based on the absorptivity of foreign matter for each energy band, using distinct processing units for images acquired in different energy bands to reflect the absorptivity accurately, allowing for the generation of suitable training data for machine learning models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If virtual defective-product images are synthesized by simply combining virtual foreign-matter images with non-defective-product images, then image generation is simplified, but the absorptivity for real electromagnetic waves is not accurately reflected
Solution Approach 1:
The patent applies parameter changes by adjusting pixel values in synthesized images based on absorptivity characteristics for different energy bands. Specifically, it modifies the synthesis process to incorporate absorptivity parameters, where pixel values are changed according to the absorptivity of foreign matter at different energy levels, thereby accurately reflecting real electromagnetic wave absorption while maintaining the virtual image generation approach
Solution Approach 2:
The patent implements local quality by applying different processing to pixels corresponding to foreign matter regions versus other regions. The system identifies target pixels that correspond to foreign matter and applies absorptivity-based pixel value changes only to these specific pixels, while leaving other pixels unchanged, thus locally enhancing the accuracy of absorptivity reflection without affecting the entire image
2Device complexity
If the same image processing is applied to both energy bands, then processing complexity is reduced, but the different absorptivity characteristics of different energy bands are not captured
Solution Approach 1:
The patent applies segmentation by dividing the image processing into distinct processing units for different energy bands. It separates the processing of first energy band images and second energy band images, with each processing unit applying appropriate absorptivity-based pixel value changes specific to its energy band, thereby capturing the different absorptivity characteristics while maintaining organized processing structure
Solution Approach 2:
The patent implements parameter changes by using different processing parameters for different energy bands. The system changes pixel values based on absorptivity parameters that are specific to each energy band, allowing the processing to reflect the unique absorption characteristics of foreign matter at different energy levels without requiring completely different processing algorithms
3Productivity
If virtual defective-product images do not accurately reflect absorptivity, then training data generation is faster, but machine learning model performance deteriorates
Solution Approach 1:
The patent applies parameter changes by incorporating absorptivity parameters into the virtual defective-product image generation process. It modifies pixel values based on the absorptivity of foreign matter for different energy bands, ensuring that the generated training data accurately reflects real electromagnetic wave absorption characteristics, thereby improving machine learning model effectiveness while maintaining efficient generation through the virtual synthesis approach
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The solution enables the generation of virtual defective-product images that accurately reflect absorptivity differences across energy bands, improving the effectiveness of machine learning models in detecting foreign matters by providing training data that accounts for varying electromagnetic wave absorption.
Implementation Method 1
a first processing unit configured to change a pixel value of a first target pixel that is at least one pixel forming the first non-defective-product image, and thereby generate a first virtual defective-product image... by first processing based on an absorptivity of the actual foreign matter for each electromagnetic wave
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An image generation device comprising: an image acquiring unit configured to acquire a set of a first non-defective-product image and a second non-defective-product image for the same article (G) to be inspected; a first processing unit configured to change a pixel value of a first target pixel that is at least one pixel forming the first non-defective-product image, thereby generating a first virtual defective-product image that is the virtual defective-product image; and a second processing unit configured to change a pixel value of a second target pixel that corresponds to the first target pixel and is at least one pixel forming the second non-defective-product image, thereby generating a second virtual defective-product image that is the virtual defective-product image.