Streak Defect Classification Using Intensity Variation Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting streak-shaped defects in inkjet printing fail to accurately classify the degree of defects, leading to inconsistent quality evaluation and ineffective maintenance strategies.
Innovation Solution
A method involving a streak-shaped defect classification system that utilizes a learned learning model to classify defects based on local maximum values and variations in average intensity values, mimicking human visual perception.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional threshold-based detection method is used, then streak-shaped defects can be detected, but the degree of defect cannot be classified
Solution Approach 1:
The patent segments the defect detection process into multiple stages: first detecting the presence of streak-shaped defects using threshold-based methods, then further analyzing the detected defects by calculating average intensity values and their variations to classify defect degrees into multiple categories (e.g., small, medium, large). This segmentation allows both detection and classification to be performed systematically.
Solution Approach 2:
The patent transitions from one-dimensional threshold comparison to two-dimensional analysis by introducing both average intensity values and their variations as classification criteria. This dimensional expansion enables the system to classify defects into multiple severity levels rather than simply detecting their presence, thereby preserving defect degree information.
2Productivity
If automatic classification based on average intensity value is used, then classification speed is improved, but classification accuracy does not match human visual observation
Solution Approach 1:
The patent changes the parameters used for classification from simple average intensity values to a combination of average intensity values and their variations. By incorporating variation as an additional parameter, the classification system achieves results that better match human visual observation while maintaining automatic processing speed.
Solution Approach 2:
The patent incorporates feedback mechanisms where classification results are continuously refined by comparing automated classifications with human visual observations. The system uses the variation in intensity values as feedback to adjust and improve classification accuracy, ensuring that automatic classification aligns with human perception standards.
3Loss of information
If detailed defect analysis is performed, then defect degree classification is achieved, but processing complexity increases
Solution Approach 1:
The patent performs preliminary defect detection using simple threshold-based methods before conducting detailed analysis. By first identifying potential defects and then focusing computational resources only on those specific regions, the system achieves detailed defect classification without proportionally increasing overall processing complexity.
Solution Approach 2:
The patent extracts only the necessary features (average intensity values and their variations) from the defect regions for classification purposes. By taking out and analyzing only the relevant characteristics rather than processing entire images in detail, the system achieves accurate defect degree classification while keeping processing complexity manageable.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
By causing a learning model to learn a relationship between feature values and classification destinations using at least a peak average intensity value and a variation in average intensity value around a streak-shaped defect as the feature values, a classification model for classifying the streak-shaped defect is generated (S150). Thereafter, when the streak-shaped defect is detected, at least the peak average intensity value and the variation are obtained as the feature amounts representing features of the streak-shaped defect (S180). Then, by inputting the feature amounts to the classification model, the classification destination depending on the features of the detected streak-shaped defect is decided (S190).