Neural Network State Decision for Image Defect Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current pattern recognition technologies, particularly in machine learning, face challenges in accurately determining the state of image data using neural networks, as they often require labeled data and struggle with unsupervised learning methods, limiting their effectiveness in identifying defects in images without prior knowledge.
Innovation Solution
A method for state decision of image data using a network function learned with respect to at least one pattern, which involves acquiring output data through a neural network and an additional algorithm, and deciding state information based on the similarity or difference between these outputs, allowing for the identification of defects in images without labeled data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If supervised learning method is used with labeled data, then pattern recognition accuracy is improved, but data preparation complexity and time consumption increase
Solution Approach 1:
The patent applies preliminary action by pre-training the neural network with labeled data to learn pattern recognition capabilities, then using the trained network for unsupervised defect detection. The network is prepared in advance with supervised learning, enabling it to subsequently identify defects without requiring labeled defect data during actual inspection operations.
Solution Approach 2:
The patent uses an intermediary approach by introducing a trained neural network as a mediator between the image data and defect identification. The network, pre-trained with labeled data, serves as an intermediary that has already learned pattern recognition, allowing the system to detect defects in new images without requiring new labeled data or complex preparation for each inspection task.
2Loss of time
If unsupervised learning method is used without labeled data, then data preparation time is reduced, but pattern recognition accuracy deteriorates
Solution Approach 1:
The patent applies preliminary action by pre-training the neural network with labeled data before deployment. This preliminary supervised training equips the network with pattern recognition capabilities, enabling it to subsequently perform unsupervised defect detection with high accuracy without requiring labeled defect data during actual operation.
Solution Approach 2:
The patent uses copying by creating a trained model that captures pattern recognition knowledge from labeled training data. This trained network serves as a copied representation of defect patterns that can be applied to multiple inspection tasks without requiring the original labeled data or repeating the training process for each new inspection.
3Measurement precision
If neural network with multiple nodes and weights is used, then pattern recognition capability is improved, but device complexity increases
Solution Approach 1:
The patent applies preliminary action by pre-configuring and training the neural network structure with appropriate nodes, links, and weights before deployment. This preliminary setup optimizes the network architecture for specific defect detection tasks, enabling complex pattern recognition capabilities to be achieved through pre-established configurations rather than complex real-time computations.
Data Source
AI summary
Disclosed is a method for state decision of image data. The method for state decision of image data may include: acquiring first output data by the network function based on the image data; acquiring second output data by an algorithm having a different effect from the network function based on the image data; and deciding state information of the image data based on the first output data and the second output data.