AVI Feedback Interface for Defect Classification Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional automated visual inspection systems for woven articles fail to achieve 100% accuracy in defect detection and classification due to insufficient training on diverse defect types, leading to misclassification and missed defects.
Innovation Solution
An interactive user feedback system implemented in an electronic device, allowing inspectors to review and correct defect classification data through a display of sub-regions with zoom-in images, and update the defect inspection model using revised data to enhance accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If an automated visual inspection system uses a pre-trained defect inspection model to automatically detect and classify defects, then inspection efficiency is improved, but inspection accuracy deteriorates due to insufficient training on diverse defect types
Solution Approach 1:
The system implements a feedback mechanism where inspectors can review defect classification results and provide corrections. The corrected data is then used to retrain and update the defect inspection model, creating a continuous improvement loop that enhances accuracy while maintaining automated inspection efficiency
Solution Approach 2:
The system performs preliminary automated defect detection and classification before human inspection, filtering out obvious cases and presenting only uncertain or complex cases to inspectors for review, thereby maintaining high efficiency while improving accuracy through targeted human feedback
2Measurement precision
If the defect inspection model is trained with more diverse defect samples to improve classification accuracy, then measurement precision is improved, but device complexity increases
Solution Approach 1:
The system enables self-service model updating where inspectors directly provide corrected defect classification data during normal operation, and the system automatically retrains the model using this feedback without requiring complex external intervention or specialized training procedures
Solution Approach 2:
Instead of requiring complete retraining with extensive diverse samples, the system uses partial updates with corrected defect data from inspectors, applying incremental learning to improve accuracy without the complexity of comprehensive model retraining
Data Source
AI summary
An interactive user feedback system for enhancing the inspection accuracy of an automated visual inspection (AVI) system is disclosed. When working normally, the AVI system acquires an article image from a continuous article transferred by a transfer equipment, and then determines whether there is at least one defect feature existing in the article image or not. Subsequently, the interactive user feedback system enables a display of the electronic device show an inspection report region consisting of M×N sub-regions. As such, the display is enabled to show a zoom-in sub-image containing at least one defect after a sub-region is clicked. Therefore, by viewing the zoom-in sub-image, an inspector can determine whether a defect classification data made by the AVI system is correct or not. If not, the inspector is able to revise the defect classification data through the interactive user feedback system.