Printed Material Inspection Using Defect-Specific Reference Preprocessing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing printed material inspection technologies face challenges in accurately and efficiently detecting multiple types of defects in printed materials due to increased processing time and complexity, particularly when inspecting a series of printed materials in a production process.
Innovation Solution
A printed material inspection device that preprocesses reference data based on defect types to generate customized comparison data, using deep learning models for feature extraction and conversion, reducing dynamic processing and improving detection accuracy and speed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single machine learning model is used to detect multiple types of defects, then the device complexity is reduced, but the measurement precision and detection accuracy deteriorate
Solution Approach 1:
The patent divides the defect detection task into multiple specialized machine learning models, each trained to detect a specific type of defect (e.g., streak defects, spot defects, irregular defects). This segmentation allows each model to focus on specific defect characteristics, thereby improving detection accuracy without requiring a single complex model to handle all defect types simultaneously.
Solution Approach 2:
Instead of training one model to detect all defect types equally well, the patent applies partial action by creating specialized models for each defect type. Each model performs excessive action on its specific target defect type, achieving superior detection performance for that particular defect category while maintaining overall system efficiency.
2Measurement precision
If complex preprocessing is performed to improve defect extraction accuracy, then the measurement precision improves, but the processing time increases
Solution Approach 1:
The patent performs preprocessing operations such as image binarization, noise filtering, and feature enhancement in advance, before the actual defect detection process. By completing these computationally intensive operations beforehand, the system reduces the processing time required during real-time inspection while maintaining high defect extraction accuracy through the use of pre-processed, optimized data.
3Measurement precision
If multiple preprocessing types are performed on reference data for different defect types, then the measurement precision improves, but the device complexity and processing overhead increase
Solution Approach 1:
The patent applies different preprocessing techniques tailored to specific defect types. For example, streak defect detection may use horizontal smoothing filters, while spot defect detection may use Gaussian blur or adaptive thresholding. This local quality approach ensures that each defect type receives the most appropriate preprocessing treatment, improving detection accuracy without requiring all preprocessing methods to be applied uniformly to all defect types, thereby reducing overall system complexity.
Data Source
AI summary
A printed material inspection device for detecting types of defects of a printed material by using inspection data based on a captured image of the printed material and reference data which is a digital image of a printing target image indicated by print data of the printed material, includes: one or more storage devices; and one or more processors configured to acquire the reference data, perform, as preprocessing of converting the reference data for comparing the reference data and the inspection data to each other, types of preprocessing, which are different depending on a defect type of a detection target, on the reference data before the inspection data is acquired, to generate pieces of comparison reference data, which are different depending on the defect type of the detection target, from the reference data, and hold the pieces of comparison reference data in the one or more storage devices.