Fusion Neural Networks for Unaligned Defect Image Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current automated defect classification systems in display panel manufacturing are inefficient due to the need for high-resolution image analysis, lack of consistency across operators, and challenges in accurately identifying defects across various image types and orientations, leading to time-consuming and inaccurate defect identification.
Innovation Solution
A system utilizing a fusion neural network that combines independently trained neural network branches, each processing unaligned data from different image sources, with attention modules to focus on defect areas and channels, enabling more accurate and human-understandable defect classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual defect classification is used, then operators can identify defect features, but the process is time-consuming and lacks consistency
Solution Approach 1:
The system segments the defect classification task into multiple independent neural network branches, each specializing in different defect types or image characteristics. This segmentation enables parallel processing of different defect features, reducing overall classification time while maintaining accuracy through specialized analysis in each branch.
Solution Approach 2:
The system introduces an intermediary attention mechanism that selectively highlights relevant defect features in input images. This attention module acts as a mediator between raw image data and classification decisions, automatically identifying and emphasizing critical defect characteristics to improve both speed and accuracy of classification.
2Measurement precision
If complex machine learning approaches are used, then defect identification accuracy can be improved, but system complexity increases
Solution Approach 1:
The complex classification problem is divided into multiple simpler sub-tasks handled by separate neural network branches. Each branch processes specific aspects of defect detection independently, reducing the complexity burden on any single component while collectively achieving high overall accuracy through their combined outputs.
Solution Approach 2:
The system transforms the classification problem by adding the dimension of multi-branch parallel processing. Instead of using a single complex network, it creates multiple simpler networks operating in parallel, each contributing to the final classification decision from different analytical perspectives, thereby managing complexity while enhancing accuracy.
3Measurement precision
If high-resolution images are analyzed, then defect detection accuracy improves, but processing speed decreases
Solution Approach 1:
The high-resolution image analysis task is segmented across multiple neural network branches that process different regions or features in parallel. This segmentation allows the system to maintain high detection accuracy by analyzing detailed image features while improving processing speed through concurrent execution of multiple analysis streams.
Solution Approach 2:
The attention mechanism applies partial action by selectively focusing computational resources only on the most relevant defect regions within high-resolution images. Instead of processing the entire high-resolution image uniformly, the system concentrates analysis on critical areas, maintaining detection accuracy while reducing overall processing time and computational load.
Data Source
AI summary
A system for manufacturing defect classification is presented. The system includes a first neural network receiving a first data as input and generating a first output, a second neural network receiving a second data as input and generating a second output, wherein first neural network and the second neural network are trained independently from each other, and a fusion neural network receiving the first output and the second output and generating a classification. The first data and the second data do not have to be aligned. Hence, the system and method of this disclosure allows various type of data that are collected during manufacturing to be used in defect classification.