Far-Infrared Image Conversion for Color-Preserving Transfer Learning

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Transfer learning using far-infrared images as training data for object detection models results in a loss of color information, leading to reduced recognition accuracy.

Innovation Solution

Convert far-infrared images into visible light images using a machine-learned image conversion model, and perform transfer learning on a visible light image trained model using the converted images as training data to generate a second visible light image trained model.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Illumination intensity

If transfer learning is performed using far-infrared images as training data, then object detection can be achieved in low-light conditions, but color information is lost and recognition accuracy decreases

Engineering Contradiction:

Improvelow-light detection capabilityVSAvoidcolor information loss

Core Design Contradiction:

Illumination intensityVSLoss of information

Solution Approach 1:

The patent introduces a visible light image as an intermediary representation. The far-infrared image is converted to a visible light image format, which then serves as training data for transfer learning. This intermediary preserves color information while enabling the use of far-infrared capture capabilities, thus resolving the contradiction between low-light detection and color information preservation

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent changes the parameter representation of the far-infrared image by converting it to visible light image format. This parameter transformation maintains the essential visual information including color, while adapting the data to be suitable for transfer learning on visible light trained models, thereby preventing information loss

Inventive Principle:

Principle #35Parameter changes

2Adaptability or versatility

If transfer learning is performed using far-infrared images, then the model can operate without visible light, but inference accuracy is reduced due to lack of color information

Engineering Contradiction:

Improveoperation without visible lightVSAvoidinference accuracy

Core Design Contradiction:

Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The visible light image serves as a mediator that bridges the gap between far-infrared capture and visible light trained models. By converting far-infrared images to visible light format, the system maintains adaptability for low-light operation while preserving the color information necessary for high inference accuracy

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent creates a copy of the far-infrared image in visible light format. This copied representation retains all color and visual information characteristics of visible light images, allowing the model to operate as if it were processing actual visible light images, thus maintaining both adaptability and accuracy

Inventive Principle:

Principle #26Copying

Data Source

PatentUS12423955B2Machine learning device and far-infrared image capturing device

Publication Date: 2025.09.23 JVC KENWOOD CORP

AI summary

A far-infrared image acquisition unit acquires a far-infrared image. An image conversion unit converts the acquired far-infrared image into a visible light image. A visible light image trained model storage unit stores a first visible light image trained model having performed learning using the visible light image as training data. A transfer learning unit performs transfer learning on a first visible light image trained model by using the visible light image obtained by conversion as training data to generate a second visible light image trained model.