Coarse-to-Fine Subtle Defect Detection via Dual-Stage Network
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional manual defect detection methods in industrial manufacturing are inefficient and prone to missing subtle defects, while machine vision-based methods fail to accurately locate and classify these defects, leading to low precision and high missing detection rates.
Innovation Solution
A coarse-to-fine defect detection method using a charge-coupled device (CCD) camera, involving a defect area location network and a defect point detection network trained with a segmentation loss function, which includes a backbone network, classification network, and regression network, to accurately locate and classify subtle defects through image preprocessing and feature extraction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If manual detection is used, then labor consumption is high and efficiency is low, but detection precision is poor due to missing subtle defects
Solution Approach 1:
The detection process is divided into two stages: a first detection model performs coarse detection to locate potential defect areas, and a second detection model performs fine detection to precisely identify subtle defects. This segmentation allows the system to achieve both high efficiency in screening and high precision in final detection.
Solution Approach 2:
The first detection model performs preliminary detection to pre-identify potential defect regions before the second model conducts detailed analysis. This preliminary action reduces the search space for the second model, improving both efficiency and precision in detecting subtle defects.
2Productivity
If machine vision-based detection is used, then detection efficiency is improved, but subtle defects cannot be accurately located and classified
Solution Approach 1:
The system uses a two-stage detection architecture where the first model provides coarse localization and the second model provides fine-grained defect classification and precise location. This segmentation enables machine vision to achieve both efficiency and precision in subtle defect detection.
Solution Approach 2:
The first detection model acts as an intermediary that processes the input image and outputs potential defect regions, which then serve as input for the second detection model. This intermediary stage enables the system to efficiently narrow down search areas before performing detailed defect analysis.
3Measurement precision
If a single detection model is used, then the system is simpler, but it cannot simultaneously achieve high efficiency and high precision in subtle defect detection
Solution Approach 1:
The detection system is segmented into two specialized models: a first model optimized for efficient coarse detection and a second model optimized for precise subtle defect identification. This segmentation allows each model to be specialized for its specific task, achieving high overall precision without excessive complexity.
Solution Approach 2:
The system transitions from a single-dimension detection approach to a two-dimensional hierarchical approach, where the first model operates at a coarse level and the second model operates at a fine level. This dimensional change enables the system to handle both efficiency and precision requirements simultaneously.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This method effectively reduces the missing detection rate and improves the accuracy and efficiency of subtle defect detection by accurately recognizing and locating micro defects, eliminating false detections and enhancing overall detection precision.
Implementation Method 1
acquiring data of an image to be detected via a charge-coupled device (CCD) camera
Data Source
AI summary
A subtle defect detection method based on coarse-to-fine strategy, including: (S1) acquiring data of an image to be detected via a charge-coupled device (CCD) camera; (S2) constructing a defect area location network and preprocessing the image to be detected to initially determine a defect position; (S3) constructing a defect point detection network; and training the defect point detection network by using a defect segmentation loss function; and (S4) subjecting subtle defects in the image to be detected to quantitative extraction and segmentation via the defect point detection network.