Saliency Map Defect Localization for AOI Image Inspection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automated optical inspection (AOI) systems struggle to accurately determine the exact position and type of defects in images, limiting the effectiveness of image classification models in manufacturing processes.
Innovation Solution
A method involving the generation of saliency maps, followed by dimensionality reduction techniques such as PCA, UMAP, or t-SNE, to spatially resolve defects in components, allowing for precise localization and clustering of defect locations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional AOI systems use deep learning-based image classification models to detect defects, then defect detection capability is improved, but the ability to determine exact defect position and type deteriorates
Solution Approach 1:
The patent segments the defect analysis process into two distinct stages: first, a deep learning classification model identifies whether a defect exists and categorizes it; second, a saliency map generation process specifically highlights the spatial location and characteristics of the detected defect. This segmentation allows each component to be optimized for its specific function, resolving the contradiction between detection capability and position determination precision.
Solution Approach 2:
The patent introduces saliency maps as an intermediary between the deep learning classification model and the final defect analysis. The saliency map serves as a mediator that translates the classification model's output into spatially-resolved information, enabling both defect detection and precise localization without requiring the classification model itself to provide position information.
2Reliability
If experts manually label images to train machine learning models, then model training quality is improved, but the labeling process focuses on defect presence rather than specific location, losing spatial information
Solution Approach 1:
The patent applies dimensionality reduction methods as a preliminary action to the saliency maps before expert review. By pre-processing the saliency maps through techniques like PCA, UMAP, or t-SNE, the system prepares the spatial information in an optimized format that highlights defect locations and patterns, making it easier for experts to review and label data with spatial context already organized.
Solution Approach 2:
The patent replaces the traditional manual labeling process with an automated saliency map generation and dimensionality reduction pipeline. Instead of experts directly labeling images (a mechanical/manual process), the system uses computational algorithms to automatically generate saliency maps and reduce their dimensions, substituting the manual mechanical labeling process with an automated computational system that preserves spatial information.
3Ease of operation
If dimensionality reduction is applied to feature vectors, then data visualization is simplified, but the ability to determine exact defect position and type is lost
Solution Approach 1:
The patent applies dimensionality reduction specifically to the saliency maps (which already represent spatial information in a condensed form) rather than to the original high-dimensional feature vectors. This approach reduces the visual complexity of the saliency maps for easier interpretation while preserving the spatial relationships encoded in the saliency map structure, thereby maintaining defect position determination capability.
Data Source
AI summary
A method for the spatially resolved localization of a defect in a component includes (i) providing a saliency map of a NOK image or a NOK defect class image of a component exhibiting a defect, (ii) applying at least one dimension reduction method to the provided saliency map to generate a dimension-reduced saliency map, and (iii) spatially resolved localization of the defect of the component on the basis of the dimension-reduced saliency map.