Line-Scan Image Feature Mapping for Annotation-Free Defect Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for defect detection in line scan images require large amounts of annotated images for training, leading to increased costs and reduced training efficiency due to the need for extensive supervised learning.
Innovation Solution
An image processing method involving downsampling of line scan images into first and second sampled images, which are input into a preset backbone network with shared feature extraction networks for self-supervised training, allowing the network to learn invariant features and context information, thereby reducing the need for sample collection and annotation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If supervised training with large amounts of annotated images is used, then detection accuracy is improved, but training cost and time increase significantly
Solution Approach 1:
The system performs self-supervised training where the backbone network generates its own training data and labels through downsampling and feature extraction, eliminating the need for external annotated images. The network uses itself as the supervisor, automatically creating training pairs from line scan images without human annotation.
Solution Approach 2:
The patent creates synthetic training data by downsampling the original line scan images to generate first and second sampled images. These copied versions serve as training data, replacing the need for real annotated images. The feature extraction network copies and transforms the original image features to create training pairs.
2Measurement precision
If supervised training with annotated images is used, then detection accuracy is improved, but training cost increases
Solution Approach 1:
The system generates its own training data through self-supervised learning, where the backbone network processes line scan images and creates training pairs automatically. This eliminates the need for external annotated training data, reducing both data collection cost and annotation labor.
Solution Approach 2:
The patent creates synthetic training images by downsampling the original line scan images. These copied and transformed images serve as the training data, replacing the need to collect and annotate large quantities of real defect images.
3Loss of information
If feature extraction network processes line scan images, then image features are extracted, but computational complexity increases
Solution Approach 1:
The patent divides the line scan image into multiple image blocks and processes them through downsampling to create first and second sampled images. This segmentation allows the feature extraction network to handle smaller, more manageable image portions, reducing the computational burden on each processing step.
Solution Approach 2:
The patent applies downsampling to reduce the resolution of image blocks before feature extraction. This partial action (reducing resolution) decreases the computational complexity while still preserving sufficient feature information for effective defect detection.
Data Source
AI summary
This application discloses an image processing method and apparatus, and a storage medium. The method includes receiving a line scan image in line scan data, the line scan image comprising a plurality of image blocks; obtaining a first sampled image and a second sampled image corresponding to each image block, inputting the first sampled image and the second sampled image into a preset backbone network, to obtain a first image feature corresponding to the first sampled image and a second image feature corresponding to the second sampled image, mapping the first image feature to a feature space of the second image feature to obtain a mapping feature; performing self-supervised training on the preset backbone network, and determining a to-be-processed image feature of a to-be-processed image through the target backbone network.