2D Image Defect Inspection via Neural Network Feature Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional 3D camera-based defect inspection is costly and time-consuming, and human inspections suffer from eye fatigue and require experienced workers, leading to inefficiencies in quality control.
Innovation Solution
A method using 2D image information processed by a neural network to inspect defects in real-time, involving image acquisition, gamma correction, feature extraction, and feature vector analysis to predict defect presence and location.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional 3D camera is used for defect inspection, then measurement precision is improved, but device cost and processing time increase
Solution Approach 1:
The patent uses 2D images as simplified copies of 3D defect information. Instead of processing full 3D data from expensive 3D cameras, the system captures multiple 2D images from different angles and extracts defect features from these 2D representations, achieving comparable inspection accuracy with significantly reduced processing time and cost
Solution Approach 2:
The patent extracts only the essential defect-related features from 2D images using neural networks. By focusing on extracting specific defect characteristics (such as edges, textures, and patterns) rather than processing complete 3D spatial data, the system achieves efficient real-time inspection without the computational burden of full 3D processing
2Measurement precision
If conventional 3D camera is used for defect inspection, then measurement precision is improved, but device cost increases
Solution Approach 1:
The patent replaces expensive 3D camera systems with inexpensive 2D cameras. Multiple low-cost 2D cameras can be positioned at different angles to capture images, and the neural network processes these 2D images to extract defect information, achieving the same inspection function at a fraction of the hardware cost
Solution Approach 2:
The system uses 2D image copies instead of 3D data. By processing 2D representations that contain sufficient defect information, the patent eliminates the need for expensive 3D sensing hardware while maintaining inspection accuracy through intelligent image analysis
3Adaptability or versatility
If human workers are used for defect inspection, then adaptability is improved, but productivity decreases due to eye fatigue and need for breaks
Solution Approach 1:
The patent implements an automated inspection system where the neural network performs defect detection independently without human intervention. The system continuously processes images and identifies defects automatically, eliminating eye fatigue and the need for breaks, thereby maintaining high productivity and consistent performance over extended periods
Solution Approach 2:
The patent replaces the human visual inspection system with an automated neural network-based system. The neural network processes images and detects defects algorithmically, substituting human eyes and brain processing with computational algorithms that do not suffer from fatigue, enabling continuous high-speed inspection
4Adaptability or versatility
If human workers are used for defect inspection, then adaptability is improved, but loss of time increases due to training requirements
Solution Approach 1:
The neural network system performs defect inspection autonomously without requiring human operators. The system is trained once on labeled defect data and then independently applies the learned patterns to inspect products, eliminating the need for continuous worker training and experience accumulation while maintaining consistent inspection quality
Data Source
AI summary
Disclosed is a method for predicting presence or absence of a defect of a product, which is performed by one or more processors. The method may include: obtaining one or more images; inputting the obtained image into a neural network model, and generating a plurality of feature maps; extracting a first feature vector based on the plurality of feature maps; extracting a second feature vector for identifying a feature map related to the presence or absence of the defect among the plurality of feature maps based on the extracted first feature vector; extracting a third feature vector for identifying a feature map region related to the presence or absence of the defect based on the plurality of feature maps; and predicting the presence or absence of the defect based on the first feature vector, the second feature vector, and the third feature vector by using the neural network model.