Line Defect Detection Through Segmented Image Annotation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image annotation techniques for training machine learning models to detect line defects on specimens are inefficient due to the length and orientation of line defects, which often span multiple fields of view and require manual pixel labeling, making the process time-consuming and strenuous.
Innovation Solution
An improved annotation method where users draw multiple annotations along the length of line defects, forming a training data set that trains machine learning models to detect these defects by recognizing individual segments, allowing for efficient classification and detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional image annotation techniques are used to manually label pixel-level line defects, then detection accuracy can be achieved, but the annotation process becomes extremely time-consuming and labor-intensive
Solution Approach 1:
The patent segments line defects into multiple short line segments instead of requiring pixel-level annotation of entire long defects. Users annotate only local segments, which are then connected to form complete defect detections. This segmentation approach maintains detection accuracy while dramatically reducing annotation time and effort.
2Reliability
If manual pixel labeling is performed for line defects spanning multiple fields of view, then comprehensive defect coverage is achieved, but the complexity and effort of annotation increases significantly
Solution Approach 1:
The system divides long line defects that span multiple fields of view into smaller manageable segments. Each segment can be annotated independently within its local field of view, reducing annotation complexity while ensuring complete defect coverage through segment connection.
Solution Approach 2:
The system introduces an intermediary processing step that connects annotated segments to form complete line defects. This intermediary connection mechanism handles the complexity of multi-field-of-view defects automatically, reducing the burden on users while maintaining detection completeness.
3Manufacturing precision
If traditional annotation methods are used for long line defects, then all defect portions can be captured, but the annotation process becomes strenuous and inefficient
Solution Approach 1:
By segmenting long line defects into shorter portions, users can annotate each segment quickly and accurately without the strenuous task of annotating entire long defects. The segmentation maintains annotation precision while improving overall productivity through reduced annotation effort.
Data Source
AI summary
A computing system generates a training data set for training a machine learning model to detect line defects on a surface of a specimen. The computing system trains the machine learning model to detect line defects based on the training data set. The machine learning model is trained to detect line defects by detecting the individual defects represented by the plurality of annotations extending between the starting point and the end point. The computing system determines that the machine learning model has exceeded a threshold level of accuracy. Based on the determining, the computing system deploys the machine learning model to detect line defects on future specimens.