Satellite Image Super Resolution via Machine Learning Upscaling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Satellite imagery faces challenges due to frequent occlusion by clouds, resulting in limited availability of high-resolution images, which are more useful but captured less frequently and at higher costs, while low-resolution images are captured more frequently but are less useful and often obscured by cloud noise.
Innovation Solution
The technology generates predicted high-resolution images using machine learning models and super-resolution techniques by processing low-resolution images, combining them with land structure features and prior high-resolution images, allowing for the creation of high-resolution images without the need for current high-resolution data from satellites.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high-resolution satellite images are captured more frequently, then image quality and detail are improved, but capture cost and resource consumption increase
Solution Approach 1:
The patent creates high-resolution image copies through super-resolution processing of low-resolution satellite images using machine learning models. Instead of capturing new high-resolution images, the system generates synthetic high-resolution versions by upscaling low-resolution inputs, thereby avoiding additional capture costs while providing high-resolution imagery for analysis.
Solution Approach 2:
The patent replaces the mechanical satellite capture system with a computational image processing system. Rather than physically deploying satellites to capture high-resolution images, the system uses neural networks and super-resolution algorithms to computationally generate high-resolution images from readily available low-resolution satellite data.
2Productivity
If low-resolution satellite images are captured more frequently, then image availability is improved, but image quality and usefulness deteriorate
Solution Approach 1:
The patent changes the resolution parameter of low-resolution satellite images through super-resolution processing. By applying machine learning-based upscaling algorithms, the system transforms images from low resolution to high resolution, thereby maintaining frequent availability while improving image quality and usefulness for various applications.
Solution Approach 2:
The patent creates high-resolution copies of frequently captured low-resolution satellite images. By generating synthetic high-resolution versions through computational processing, the system enables frequent monitoring and analysis with high image quality without requiring frequent high-resolution satellite passes.
3Measurement precision
If super-resolution processing is applied to low-resolution satellite images, then high-resolution image availability is improved, but processing complexity and computational requirements increase
Solution Approach 1:
The patent segments the super-resolution processing into multiple specialized neural network components, including sensor transformation models, land structure extraction modules, and multi-scale feature processing networks. This modular architecture divides the complex processing task into manageable segments that can be executed efficiently and independently.
Solution Approach 2:
The patent performs preliminary processing steps before main super-resolution processing, including sensor calibration, land structure feature extraction, and image alignment. By preparing the input data in advance with these preliminary actions, the main processing pipeline becomes more efficient and requires less computational complexity during the actual upscaling operation.
Data Source
AI summary
Systems and methods for generating predicted high-resolution images from low-resolution images. To generate the predicted high-resolution images, the present technology may utilize machine learning models and super resolution models in a series of processes. For instance, the low-resolution images may undergo a sensor transformation based on processing by a machine learning model. The low-resolution images may also be combined with land structure features and/or prior high-resolution images to form an augmented input that is processed by a super resolution model to generate an initial predicted high-resolution image. The predicted initial high-resolution image may be combined or stacked with other predicted high-resolution images to form a stacked image. That stacked image may then be processed by another super resolution model to generate a final predicted high-resolution image.