Thermal Image Resolution via Machine Learning Copying
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing panorama composition technique for obtaining high-resolution thermal images using low-resolution thermal cameras is inefficient, leading to prolonged image capturing times and reduced accuracy due to the lengthened process, which can result in inaccurate training data for machine learning models.
Innovation Solution
A method involving a machine learning model that converts high-resolution visible light images into high-resolution thermal images, using optical magnification to capture specific areas of uncertainty, thereby shortening the time required to obtain high-resolution thermal images while using inexpensive low-resolution thermal cameras.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple low-resolution thermal images are captured and composited into a high-resolution thermal image using panorama composition technique, then high-resolution thermal images can be obtained using inexpensive low-resolution thermal cameras, but the image capturing time is prolonged and accuracy is reduced
Solution Approach 1:
The patent uses a visible light image (first image) as a reference to generate a synthetic thermal image (second invisible light image) through machine learning, rather than capturing multiple actual thermal images. This copying approach allows obtaining high-resolution thermal images without the time penalty of multiple captures and compositing operations
Solution Approach 2:
The patent replaces the mechanical panorama composition process (capturing multiple images and compositing them) with a machine learning-based image generation system. The machine learning model directly generates high-resolution thermal images from visible light images, eliminating the need for multiple captures and mechanical compositing operations
2Measurement precision
If multiple low-resolution thermal images are captured and composited into a high-resolution thermal image, then high-resolution thermal images can be obtained using inexpensive low-resolution thermal cameras, but the accuracy of training data is reduced
Solution Approach 1:
The patent generates synthetic thermal images by copying and transforming visible light image information through a trained machine learning model. This approach produces high-resolution thermal images that maintain accuracy suitable for training data, avoiding the degradation that occurs with panorama composition of multiple low-resolution captures
Solution Approach 2:
The patent replaces the mechanical compositing process with machine learning-based image generation. The machine learning model, trained on paired visible light and thermal images, generates accurate thermal images that preserve the quality needed for reliable training data, eliminating accuracy loss associated with traditional compositing methods
3Measurement precision
If high-resolution thermal cameras are used to obtain high-resolution thermal images, then accurate training data can be obtained, but the cost increases
Solution Approach 1:
The patent uses visible light images (captured by inexpensive cameras) as references to generate synthetic high-resolution thermal images through machine learning. This copying approach eliminates the need for expensive high-resolution thermal cameras while maintaining the ability to produce accurate thermal images for training purposes
Solution Approach 2:
The patent replaces expensive high-resolution thermal cameras with inexpensive low-resolution thermal cameras or even visible light cameras. The machine learning model compensates for the lower hardware resolution, enabling high-resolution thermal image generation using cheap imaging devices
4Ease of manufacture
If panorama composition technique is used to obtain high-resolution thermal images, then inexpensive low-resolution thermal cameras can be used, but the process is inefficient and time-consuming
Solution Approach 1:
The patent replaces the multi-step mechanical panorama composition process with a single-step machine learning-based image generation process. The machine learning model directly generates high-resolution thermal images from visible light images, eliminating the need for multiple captures, alignment, and compositing operations, thereby dramatically improving efficiency
Solution Approach 2:
The patent uses visible light images as references to copy and transform thermal information through machine learning, generating high-resolution thermal images in a single operation. This approach maintains the ability to use inexpensive cameras while achieving high productivity through direct synthetic image generation
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach allows for efficient training of machine learning models with high-resolution thermal images, reducing the time and cost associated with using high-resolution thermal cameras and improving the accuracy of the training data.
Implementation Method 1
generating a second invisible light image at a resolution higher than a resolution of the first invisible light image by a machine learning model using the first image and the first invisible light image as an input
Implementation Method 2
obtaining, by an optical magnification control of the invisible light image capturing device, a third invisible light image of the obtaining target area at a resolution higher than a resolution of the obtaining target area in the first invisible light image
Data Source
AI summary
A control method including: obtaining an image of a given range captured by an image capturing device and a first invisible light image of the range captured by an invisible light image capturing device having a resolution lower than a resolution of the image capturing device; generating a second invisible light image at a resolution higher than a resolution of the first invisible light image by a machine learning model using the image and the first invisible light image as an input; identifying a target area from the range, based on an indicator indicating an uncertainty of each pixel in the second invisible light image; and obtaining, by an optical magnification control of the invisible light image capturing device, a third invisible light image of the target area at a resolution higher than a resolution of the target area in the first invisible light image.