GAN-Based Defect Data Generation for Printed Matter Inspection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current defect detection methods for printed matter are labor-intensive and require extensive labeled training data, leading to inefficiencies and inaccuracies due to the small proportion of defects in images, which can result in poor data quality and increased labor costs for manual quality checking.
Innovation Solution
A method involving the creation of a defects library, training a machine learning algorithm, and using it to detect defects in printed images by generating and validating defect datasets, ensuring data quality through principles like color difference, shape analysis, and text overlap verification, allowing for semi-supervised deep learning model training.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual quality checking is used to detect defects in printed matter, then detection accuracy can be maintained, but labor costs and time consumption increase significantly
Solution Approach 1:
The patent replaces manual mechanical inspection with an automated machine learning-based inspection system. The system uses trained ML models to automatically detect defects in printed matter, substituting human labor with computational algorithms that can process images rapidly and consistently without fatigue or distraction.
Solution Approach 2:
The inspection system is designed to be self-sufficient by automatically generating synthetic defect data through GANs and performing self-training of ML models. The system can autonomously improve its detection capabilities without requiring continuous manual intervention for data labeling or model retraining, enabling sustained high-performance operation.
2Measurement precision
If extensive labeled training data is collected for ML model training, then model accuracy improves, but data collection time and labor costs increase
Solution Approach 1:
The system performs preliminary action by pre-training ML models using synthetically generated defect data before deploying them for actual defect detection. The GAN-based data generation creates realistic defect samples in advance, allowing the model to be pre-trained and ready for production use without requiring extensive collection and labeling of real defect data at deployment time.
Solution Approach 2:
The patent uses GANs to copy and synthesize defect data that mimics real defect characteristics. The generative model creates synthetic images with various defects (stains, scratches, misalignments) that replicate the statistical properties and visual characteristics of actual defects, providing sufficient training material without requiring extensive real defect samples.
3Reliability
If the proportion of defects in training images is small, then real-world scenarios are accurately represented, but data quality decreases and detection sensitivity reduces
Solution Approach 1:
The system applies local quality by concentrating defect samples in specific regions of training images while maintaining realistic overall image distributions. The GANs generate images where defects are localized to specific areas with appropriate spatial frequencies and intensities, allowing the model to learn defect characteristics intensely in local regions while preserving the natural low defect-rate context of real printed matter.
Solution Approach 2:
The patent employs parameter changes by adjusting the defect injection parameters in synthetic data generation. The system varies defect density, size, type, and distribution parameters during training to expose the model to a wide range of defect scenarios. This allows the model to learn sensitive defect detection across different parameter regimes while maintaining realistic representations when defect proportions are low.
Data Source
AI summary
Technology for inspection for detecting a defect of a printed matter using machine logic informed by machine learning. Some embodiments of the present invention may include one, or more, of the following features: (i) generates defect datasets; (ii) generates defect libraries; (iii) uses the generated defect libraries for deep learning training; and (iv) uses machine learning to detect defects using computer code (for example, a *.jpg format file) corresponding to an image of a piece of printed matter instead of using a visual image (that is, an image of the type that is created when a person takes a picture using a traditional film camera).