Image Annotation Standardization for Defect Pattern Consistency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing artificial intelligence machine-learning systems for image recognition in industrial production lack complete satisfaction due to inconsistencies in image annotation, leading to inaccurate knowledge transfer and decision-making replacement of human thinking.
Innovation Solution
A method for standardizing image annotation using a convolution neural network, which involves receiving a defect pattern, generating and comparing judgement results, and updating the defect pattern to standardize it, ensuring consistent image classification and improving recognition accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If professional marking technicians perform data marking manually, then annotation accuracy is improved, but productivity deteriorates due to huge amounts of data
Solution Approach 1:
The patent introduces an image recognition model as an intermediary between raw images and final annotations. The model pre-processes and generates initial annotation results, which are then refined by technicians. This mediator handles the bulk of data processing, freeing technicians to focus on quality control and complex cases, thus resolving the contradiction between accuracy and productivity.
Solution Approach 2:
The system performs preliminary annotation using the image recognition model before human technicians review the data. This preliminary action generates draft annotations that technicians can quickly verify or correct, rather than creating annotations from scratch. This approach significantly improves productivity while maintaining accuracy through the review process.
2Loss of time
If image recognition models are trained with inconsistent annotations, then training speed is improved, but reliability deteriorates due to knowledge transfer failure
Solution Approach 1:
The patent implements a feedback mechanism where annotation results are reviewed and corrected by professional technicians, and these corrected annotations are used to retrain and refine the image recognition model. This continuous feedback loop ensures that inconsistent annotations are identified and corrected, improving the reliability of knowledge transfer while maintaining efficient training cycles.
Solution Approach 2:
The annotation standardization method is dynamically updated based on comparison results between different technicians' annotations. The defect pattern library evolves and adapts as more data is processed and reviewed, allowing the system to improve its annotation consistency over time rather than relying on static, potentially inconsistent initial annotations.
3Manufacturing precision
If defect patterns are frequently updated to improve accuracy, then manufacturing precision is improved, but device complexity increases due to version management
Solution Approach 1:
The patent creates standardized defect pattern templates that can be copied and reused across different product types and inspection scenarios. When new defect patterns are identified, they are added to the standardized library rather than creating unique patterns for each case. This copying approach maintains high detection accuracy while simplifying version management through a centralized, standardized library.
Solution Approach 2:
The defect pattern library is designed to be universal and multi-functional, accommodating various product types and defect categories through a standardized framework. This universality allows the same pattern management system to serve multiple purposes and product lines, reducing overall system complexity while maintaining the ability to achieve high manufacturing precision across different applications.
Data Source
AI summary
A method for standardizing image annotation includes receiving a defect pattern; marking an image according to the defect pattern to generate a first judgement result; marking the image according to the defect pattern to generate a second judgement result; comparing the first judgement result and the second judgement result to obtain a comparison result; and updating the defect pattern according to the comparison result to standardize the defect pattern. The method for standardizing image annotation of the present specification can improve the marking stability of the training data of a trained image recognition algorithm, thereby improving the accuracy of image recognition of the trained image recognition algorithm.