Multi-Scale Gridding for Fast Copper-Clad Laminate Defect Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional quality inspection methods for copper-clad laminates are inefficient, labor-intensive, prone to missing defects due to human perception limitations, and time-consuming, especially when dealing with large materials, and existing visual detection methods struggle with detecting defects of varying sizes and are slow.
Innovation Solution
A method utilizing multi-scale gridding and deep learning-based detection, involving image segmentation, pyramid operations, and a MobileNetV2 model for real-time defect detection, which includes threshold-based segmentation, affine transformation, and a feature pyramid structure for accurate defect classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual inspection is used to detect defects on copper-clad laminate, then human perception can identify some defects, but the inspection process is time-consuming and labor-intensive
Solution Approach 1:
The patent replaces manual mechanical inspection with an automated visual inspection system using a camera to capture images of the copper-clad laminate surface. The system processes images through computer algorithms including background subtraction, contour detection, and defect classification to automatically identify and categorize defects such as scratches, dirt, black spots, white spots, and bubbles, eliminating the need for manual rotation and visual inspection while maintaining high detection accuracy.
Solution Approach 2:
The patent creates a digital copy (image) of the copper-clad laminate surface and processes this copy through computational methods. By working with image data rather than physical material, the system can analyze the entire surface simultaneously without the time-consuming manual rotation process, while the digital nature of the copy allows for rapid processing and repeated analysis without additional time cost.
2Reliability
If conventional visual detection with sliding window average calculation is used, then defects can be detected through grayscale difference, but the detection process is excessively time-consuming
Solution Approach 1:
The patent divides the image processing task into distinct segmentation steps: first separating the copper-clad region from the non-copper-clad region, then dividing the copper-clad region into a grid of sub-regions. This segmentation allows the system to process only relevant areas with appropriate algorithms for each region type, avoiding the time-consuming pixel-by-pixel sliding window approach while maintaining defect detection capability through contour analysis and background subtraction.
Solution Approach 2:
The patent changes the detection parameters and methods by replacing the sliding window average calculation with a background subtraction approach using grid-based contour analysis. Instead of calculating averages across overlapping windows, the system establishes a background model from the copper-clad region and subtracts it from the current image, significantly reducing computational complexity and processing time while maintaining sensitivity to grayscale differences caused by defects.
3Measurement precision
If average calculation is performed over entire local regions, then background can be obtained, but defects occupying large proportions of regions cannot be detected due to grayscale filling
Solution Approach 1:
The patent applies local quality by treating different regions of the image with different processing approaches. The copper-clad region is segmented and processed with contour-based background subtraction, while the non-copper-clad region is handled separately. Within the copper-clad region, the grid division allows each sub-region to have its own background model, enabling accurate defect detection even when defects occupy significant portions of local areas, as the contour-based method preserves edge information better than average calculation.
4Reliability
If workers manually rotate and turn over large materials to search for defects, then the entire surface can be inspected, but the material is damaged by these operations
Solution Approach 1:
The patent replaces the mechanical operation of rotating and turning over large copper-clad laminate materials with a stationary camera-based imaging system. The camera captures images of the material surface, and the system processes these images to detect defects across the entire surface area. This substitution eliminates all mechanical contact and manipulation that could cause damage to the material, while still achieving complete surface coverage inspection through digital imaging and processing.
Data Source
AI summary
Provided is a method for detecting a surface defect of a copper-clad laminate based on multi-scale gridding. The method includes: S1, collecting a photographed image of the copper-clad laminate through a line-scan camera; S2, segmenting out a copper-clad region and a non-copper-clad region, and rotating the copper-clad region upright; S3, dividing the copper-clad region of the image of the copper-clad laminate, and performing a pyramid operation on a divided original grid image; S4, determining averages of pixel values at all levels; S5, fusing the average at all levels to obtain a final background average; and S6, searching for a final defect according to a difference between a background and a foreground.