Surface Defect Image Extraction for Variable-Size Steel Sheet Defects
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing deep learning-based methods for detecting surface defects in cold-rolled steel sheets face challenges due to varying sizes and aspect ratios of defect images, leading to feature loss and reduced classification accuracy, especially when dealing with diverse defect characteristics from different manufacturing processes.
Innovation Solution
A method involving the generation of defect matrices and scores to standardize defect images, including first, second, and third defect matrices, with standard deviation calculations and boundary detection, to extract defect images effectively, minimizing feature loss and resizing impacts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Shape
If conventional preprocessing methods (crop, resize, padding) are used to standardize defect images, then image size consistency is improved, but image features are lost and classification reliability is degraded
Solution Approach 1:
The patent segments the defect detection process into multiple stages: first detecting defect areas and generating defect matrices at different resolutions, then selectively using these matrices for classification. This segmentation allows the system to handle images of various sizes without forcing all images into a single standardized format, thereby preserving feature information while achieving size consistency where needed.
Solution Approach 2:
The patent dynamically adjusts the processing approach based on the input image characteristics. Instead of applying a fixed preprocessing pipeline, the system adapts by selecting appropriate defect matrices from different resolutions based on the detected defect areas, allowing the classification model to receive optimally processed images that preserve relevant features.
2Productivity
If defect images from various manufacturing processes are processed using fixed preprocessing methods, then processing efficiency is improved, but classification accuracy for diverse defect characteristics is degraded
Solution Approach 1:
The patent creates a universal defect detection system that can handle multiple manufacturing processes and defect types through a single framework. The system generates defect matrices at multiple resolutions and uses these universal representations for classification, allowing it to adapt to diverse defect characteristics without requiring process-specific preprocessing pipelines.
Solution Approach 2:
The patent changes the parameter of image resolution by generating defect matrices at multiple different resolutions. This allows the system to capture defect characteristics at various scales, improving classification accuracy for diverse defect types while maintaining processing efficiency through automated multi-resolution generation.
Data Source
AI summary
Provided is a method of detecting a surface defect. The method includes acquiring a target image, detecting at least one defect area from the target image, generating a first defect matrix having a size corresponding to the target image and an element that is a first defect score calculated based on the number of defect areas, calculating a plurality of second defect scores by summing first defect scores in a range corresponding to a predetermined reference window in the first defect matrix, and extracting an image corresponding to the reference window from the target image based on the second defect score and generating a defect image.