Multi-Scale Image Judgment for Accurate Substrate Inspection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image judgment systems for inspecting semiconductor and printed substrates suffer from insufficient judgment accuracy, leading to increased human resource costs and time due to reliance on skilled experts, and existing AI technologies have not adequately addressed this issue.
Innovation Solution
An image judgment apparatus and method that utilizes a judgment model learning from reference data to segment images at multiple magnifications, generating image data for input to the model to accurately classify image quality based on predefined judgment items, thereby improving inspection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If conventional AI classification methods are used for substrate inspection, then automation is achieved, but judgment accuracy deteriorates
Solution Approach 1:
The inspection system segments the substrate surface into multiple regions and performs classification judgment on each region separately. This allows the system to maintain automation while improving accuracy by analyzing local characteristics of different areas independently, rather than treating the entire substrate as a single unit.
Solution Approach 2:
The system introduces a new dimension of analysis by performing classification judgments at multiple magnification levels. By examining image data at different scales (first magnification for overall assessment, second magnification for detailed analysis), the system achieves both automation and high accuracy by capturing features that are not visible at a single scale.
2Adaptability or versatility
If multiple judgment items are used for inspection, then comprehensiveness of quality assessment is improved, but the number of wrong judgments increases
Solution Approach 1:
The system divides multiple judgment items into separate classification judgments, each targeting specific quality aspects. By segmenting the inspection into multiple specialized classification tasks rather than one comprehensive task, the system maintains comprehensiveness while reducing errors through focused analysis of each quality dimension.
Solution Approach 2:
The system adds a magnification dimension to handle multiple judgment items effectively. By performing classification judgments at different magnification levels for different judgment items, the system can comprehensively assess quality while maintaining accuracy through scale-appropriate analysis for each specific quality criterion.
3Productivity
If image data is analyzed without segmentation, then processing speed is maintained, but identification of deficiencies becomes less accurate
Solution Approach 1:
The system segments image data into multiple classification judgment targets based on different regions and magnification levels. This segmentation enables parallel processing of different image portions, maintaining overall processing speed while significantly improving deficiency identification accuracy through focused analysis of segmented regions.
Solution Approach 2:
The system introduces magnification as an additional dimension for organizing image data analysis. By processing image data at multiple magnification levels simultaneously, the system maintains productivity through efficient multi-scale analysis while achieving superior deficiency identification accuracy by examining features at appropriate scales.
Data Source
AI summary
An image judgment apparatus stores, in a memory, a judgment model configured to learn about each of the plurality of judgment items based on first data corresponding to an image as a reference with which to judge good and second data corresponding to an image as a reference with which to judge not good, sets a plurality of magnifications for each of the plurality of judgment items, segments part of the image, generates image data at the plurality of magnifications from the segmented image, inputs the generated image data corresponding to each of the plurality of magnifications to the judgment model, and judges, based on output results from the judgment model, whether the quality of the image is good or not good with respect to each of the plurality of judgment items.