X-Ray Imaging Teacher Data Segmentation for Accurate Defect Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for producing teacher data for learned models in X-ray imaging systems burden operators with manual image processing, particularly in detecting areas of inspection targets and defect parts.
Innovation Solution
An X-ray imaging system that acquires first and second images from a teacher X-ray image with regular arrangements, using a cut-out image acquirer to extract specific parts and discrimination information to reduce operator burden, and performs machine learning to produce a learned model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If operators manually apply processing to entire original images to produce teacher data, then comprehensive coverage of the image is achieved, but operator burden increases significantly
Solution Approach 1:
The patent divides the original image into multiple divided images (first image, second image, etc.) corresponding to different regions. Operators only need to manually process each divided image separately rather than the entire original image, significantly reducing the operator burden while maintaining comprehensive coverage through subsequent synthesis of results.
2Ease of operation
If divided images are used for machine learning instead of entire original images, then operator burden is reduced, but discrimination accuracy may be compromised
Solution Approach 1:
The patent introduces a synthesis processing unit that acts as an intermediary to combine the results from multiple divided images. The synthesis unit integrates the discrimination results from each divided image to produce a comprehensive analysis of the entire original image, thereby maintaining discrimination accuracy despite using divided images for processing.
3Reliability
If entire original images are processed to ensure complete analysis, then discrimination completeness is maintained, but processing time increases
Solution Approach 1:
The patent segments the original image into multiple divided images that can be processed independently and in parallel. This segmentation allows the system to maintain complete discrimination coverage while reducing overall processing time by distributing the computational workload across multiple smaller image processing tasks.
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
Reduces operator burden in teacher data production while maintaining accuracy by using cut-out images with common arrangement relations, preventing accuracy reduction in discrimination results.
Implementation Method 1
an X-ray emitter configured to emit X-rays to an inspection target having a regular arrangement; an X-ray detector configured to detect the X-rays emitted from the X-ray emitter
Data Source
AI summary
An X-ray imaging system is configured to acquire first and second images from a teacher X-ray image including an inspection target. Discrimination information to discriminate at least one of an area of the inspection target in the first and second images, and an area of a defect part is acquired. Machine learning for producing a learned model is performed by using input teacher data sets based on the first and second images, and output teacher data sets based on the discrimination information.