Time-Series Image Encoding for Correlation Change Detection

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing methods for abnormality detection in multi-dimensional time-series data, particularly using deep learning, face challenges with high training times and inefficiencies in detecting changes in correlation relationships between dimensions.

Innovation Solution

Generate a multi-dimensional first image representing frequency characteristics using wavelet transform and principal component analysis, followed by combining these images with a random matrix to create a single RGB image, which highlights changes in correlation relationships.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If deep learning is used for abnormality detection in multi-dimensional time-series data, then detection accuracy is improved, but training time becomes excessively long

Engineering Contradiction:

Improveabnormality detection accuracyVSAvoidtraining time

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The patent uses a pre-trained base model (image classifier) that was trained on large-scale natural image data instead of training a custom deep learning model from scratch for each abnormality detection task. This allows the system to achieve high detection accuracy without incurring the high training time and computational cost associated with training specialized deep learning models for each specific task.

Inventive Principle:

Principle #27Cheap short-living objects (Disposable)

2Adaptability or versatility

If multi-dimensional time-series data is imaged using wavelet transform or Gramian angular field, then the data can be processed by image classifiers, but changes in correlation relationships between dimensions are not detected

Engineering Contradiction:

Improvecompatibility with image classifiersVSAvoiddetection of correlation changes

Core Design Contradiction:

Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The patent transforms multi-dimensional time-series data into a two-dimensional image representation where one dimension represents time and the other represents frequency. This transformation using wavelet transform allows the data to be processed by image classifiers while preserving the correlation relationships between different dimensions of the original time-series data.

Inventive Principle:

Principle #17Another dimension (Dimensionality change)

3Device complexity

If individual dimensions of time-series data are imaged separately and arranged into one image, then the data structure is simplified, but correlation relationships between dimensions are lost

Engineering Contradiction:

Improvedata structure complexityVSAvoidcorrelation relationships

Core Design Contradiction:

Device complexityVSLoss of information

Solution Approach 1:

The patent merges the information from multiple dimensions of time-series data into a single two-dimensional image representation. By combining the time and frequency dimensions into one unified image structure, the system maintains the correlation relationships between dimensions while simplifying the overall data structure for processing by image classifiers.

Inventive Principle:

Principle #5Merging (Combining)

Data Source

PatentUS20250363685A1Storage medium storing image generation program, method, and device

Publication Date: 2025.11.27 FUJITSU LTD

AI summary

An image generation device includes a processor that executes a procedure. The procedure includes: generating a multi-dimensional first image representing a frequency characteristic at each time of each piece of time-series data, based on each piece of multi-dimensional time-series data; and generating a single second image obtained by combining the multi-dimensional first images weighted using a random matrix in which a different value is assigned for each frequency.