Article Inspection Image Generation for Defect Model Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing article inspection devices require a large number of images for learning due to variations in product shapes and types of defects, especially for unpredictable foreign objects and cooked/stuffed foods, making it difficult to set accurate inspection conditions without extensive image acquisition efforts.
Innovation Solution
An article inspection device uses a generative AI to create pseudo-images simulating diverse inspection scenarios, allowing for data augmentation and accurate model learning without the need for extensive image acquisition, by generating simulated images considering unpredictable variations in shape, size, and disposition of products and foreign objects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a large number of images are acquired for learning to cover variations in product shapes and types of defects, then the accuracy of inspection determination is improved, but the time and effort required for image acquisition and data preparation increases significantly
Solution Approach 1:
The patent uses generative AI to create synthetic pseudo-images that copy and simulate real inspection images with various defects and product variations. Instead of acquiring numerous real images through time-consuming processes, the system generates synthetic copies that preserve the essential characteristics needed for training the inspection determination model, thereby reducing image acquisition time while maintaining training accuracy.
Solution Approach 2:
The system performs preliminary generation of diverse training images using generative AI before the actual inspection determination process. By pre-generating a comprehensive set of pseudo-images covering various product shapes, sizes, and defect types, the system prepares the training data in advance, eliminating the need for time-consuming image acquisition during deployment and enabling rapid model training.
2Adaptability or versatility
If diverse product variations and defect types are covered in training images, then the adaptability of the inspection model is improved, but the complexity of data collection and processing increases
Solution Approach 1:
The generative AI creates synthetic copies of inspection images with controlled variations in product shapes, sizes, and defect types. This copying approach provides diverse training data without requiring complex real-world data collection processes, as the synthetic images can be generated systematically to cover all necessary variation scenarios.
Solution Approach 2:
The system varies parameters such as product shape, size, and defect characteristics in the generative AI model to produce diverse pseudo-images. By controlling these parameters during image generation, the system achieves high adaptability across different product types and defect scenarios while maintaining simple and systematic data processing, as the parameter variations are built into the generation process rather than requiring complex external data collection.
3Measurement precision
If real defective product images are used for learning, then the realism and accuracy of defect detection is improved, but the availability of sufficient training data decreases due to the rarity of actual defects
Solution Approach 1:
The system uses generative AI to create synthetic copies of defective product images, preserving the realistic characteristics of actual defects while multiplying the available training data. This copying approach generates numerous varied instances of rare defects that would be impossible to collect from actual production, thereby maintaining detection accuracy while solving the data scarcity problem.
Solution Approach 2:
The generative AI model serves itself by automatically generating the training data it needs without external intervention. Instead of relying on external sources to provide rare defective images, the system uses its own capabilities to create synthetic defect images, enabling it to train on sufficient data while maintaining the realism needed for accurate defect detection.
Data Source
AI summary
An article inspection device capable of increasing accuracy of learning or performance verification without an effort to acquire an inspection image to be used for learning or performance verification for setting an inspection condition is provided. An article inspection device includes an inspection unit that inspects a quality state of an article using an inspection image obtained by imaging the article being transported, in which the inspection unit sets an inspection condition of the quality state of the article based on a pseudo-image of the inspection image generated by a generative AI. The inspection unit inspects the quality state of the article by applying a trained model as an inspection condition created by learning the pseudo-image.