Neural Network Embedding for Anomaly Detection in Image Streams
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
High-dimensional image data introduces noise and makes it difficult to manually and automatically detect anomalies in manufactured goods, as the large amount of data and dimensionality hinder effective visualization and classification.
Innovation Solution
A neural network is trained to embed high-dimensional input data into a low-dimensional space while preserving neighbor relationships, allowing for accurate visualization and automatic classification of anomalies by grouping similar images together and reducing noise.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high-dimensional image data is used for quality control, then detailed information is preserved, but noise increases and anomaly detection becomes difficult
Solution Approach 1:
The patent extracts only the essential features from high-dimensional image data by projecting them into a low-dimensional embedding space. This extraction process removes redundant and noisy information while preserving the critical characteristics needed for anomaly detection, directly resolving the contradiction between maintaining measurement precision and reducing noise.
Solution Approach 2:
The patent introduces an embedding layer as an intermediary between the high-dimensional input data and the anomaly detection process. This embedding layer transforms the data into a lower-dimensional space that serves as a mediator, filtering out noise while preserving the essential information needed for accurate anomaly detection.
2Loss of information
If high-dimensional image data is used, then comprehensive quality information is captured, but manual identification and visualization become difficult
Solution Approach 1:
The patent transforms the data from high-dimensional space to low-dimensional space through embedding, changing the dimensionality while preserving the essential quality information. This dimensional transformation makes the data suitable for visualization and manual identification without losing critical quality characteristics.
Solution Approach 2:
The patent creates a simplified copy of the high-dimensional data in the form of low-dimensional embeddings. These embeddings serve as a compressed representation that retains the essential quality information while being much more suitable for visualization and manual analysis.
3Productivity
If automated anomaly detection is implemented, then productivity increases, but high dimensionality introduces noise that reduces detection accuracy
Solution Approach 1:
The patent extracts the essential features from high-dimensional data through embedding, removing the noisy components that would otherwise reduce detection accuracy. This allows automated systems to maintain high productivity while operating on cleaner, lower-dimensional data that improves measurement precision.
Solution Approach 2:
The embedding layer acts as an intermediary that processes high-dimensional data into a cleaner, lower-dimensional representation. This intermediary step enables automated anomaly detection to operate efficiently with improved accuracy by filtering out noise before the actual detection process.
Data Source
AI summary
Methods and systems for detecting and correcting anomalous inputs include training a neural network to embed high-dimensional input data into a low-dimensional space with an embedding that preserves neighbor relationships. Input data items are embedded into the low-dimensional space to form respective low-dimensional codes. An anomaly is determined among the high-dimensional input data based on the low-dimensional codes. The anomaly is corrected.