Probe Model Training for Repeating Pattern Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine vision systems face challenges in accurately identifying the pose of patterns with repeating elements, as current model training processes often generate non-robust models, leading to misalignment during pattern recognition, especially when patterns undergo sub-pixel phase shifts or have repeating features like a ball grid array.
Innovation Solution
A method is developed to select optimal coarse image resolutions and scan codes for training probe models, ensuring accurate pattern location by iteratively testing different resolutions and scan codes, and storing the most robust combinations for subsequent use, which allows for precise pattern recognition even in the presence of repeating elements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If known model training processes use training images to generate models at different resolutions and select maximum coarse image resolution based on circular moment of inertia, then the model development process is automated, but the resulting models are non-robust for patterns with repeating elements and lead to misalignment during recognition
Solution Approach 1:
The patent implements a feedback mechanism where the trained model is tested on the training image at multiple candidate resolutions, and the resolution that produces the most accurate pose estimates is selected. This feedback loop ensures that the chosen resolution actually improves model robustness rather than just following a predetermined selection rule.
Solution Approach 2:
The patent systematically varies the image resolution parameter during model training by generating training images at multiple candidate resolutions and testing model performance at each. This parameter exploration allows the system to identify the optimal resolution for robust pattern recognition, particularly for repeating elements.
2Ease of manufacture
If the training algorithm selects an initial image resolution that works well for aligned patterns, then the model training is simplified, but the pattern pose cannot be precisely determined when the pattern is misaligned due to sub-pixel phase shift
Solution Approach 1:
The patent performs preliminary testing of multiple candidate resolutions before final model selection. By pre-evaluating model performance at different resolutions on the training image, the system identifies the resolution that maintains accuracy under misalignment conditions, preventing future recognition failures.
Solution Approach 2:
The system uses feedback from testing the model on the training image to determine which resolution produces the most robust pose estimates. This feedback ensures the selected resolution can handle misalignment and sub-pixel phase shifts that occur during actual pattern recognition.
3Productivity
If multiple different resolutions are used to iteratively determine pattern pose, then the pattern recognition speed is improved, but the computation time increases due to testing multiple resolutions during training
Solution Approach 1:
The patent tests a limited set of candidate resolutions rather than all possible resolutions. By selecting a manageable number of candidate resolutions for training and testing, the system achieves sufficient model robustness without excessive computational time during training.
Solution Approach 2:
The patent efficiently explores the resolution parameter space by generating training images at multiple candidate resolutions and systematically testing model performance. This structured parameter exploration finds optimal resolutions without exhaustive computation.
4Measurement precision
If the model is trained at higher resolution to capture fine details, then the pattern recognition accuracy is improved, but the model becomes less robust to sub-pixel phase shifts and misalignment
Solution Approach 1:
The patent systematically varies the training image resolution parameter and tests model performance at each candidate resolution. This parameter exploration identifies the optimal resolution that balances fine detail capture with robustness to sub-pixel phase shifts and misalignment.
Solution Approach 2:
The system uses feedback from testing the trained model on the training image to evaluate robustness. By measuring pose estimation accuracy at different resolutions, the system identifies which resolution provides the most reliable performance under misalignment conditions.
Data Source
AI summary
A method for training a pattern recognition algorithm including the steps of identifying the known location of the pattern that includes repeating elements within a fine resolution image, using the fine resolution image to train a model associated with the fine image, using the model to examine the fine image resolution image to generate a score space, examining the score space to identify a repeating pattern frequency, using a coarse image that is coarser than the finest image resolution image to train a model associated with the coarse image, using the model associated with the coarse image to examine the coarse image thereby generating a location error, comparing the location error to the repeating pattern frequency and determining if the coarse image resolution is suitable for locating the pattern within a fraction of one pitch of the repeating elements.