Multi-Band Image Generation Using Deep Learning for Missing Bands
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional satellite image systems face challenges with increased size, weight, and cost due to multiple image sensors, and are prone to band failures or noise during missions, with no effective solutions for these issues.
Innovation Solution
A deep-learning network-based system that generates multi-band images by fusing data from different sensors, using training data sets to create a model that can generate missing band data, reducing the need for physical sensors and improving reliability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If multiple image sensors are mounted to capture different wavelength bands, then image information completeness is improved, but device size, weight, and cost increase
Solution Approach 1:
The patent uses deep learning models to generate synthetic image data that copies and replicates the information that would be captured by physical sensors. The model learns from training data and synthesizes missing band images, effectively replacing the need for actual sensor hardware while maintaining information completeness.
Solution Approach 2:
The deep learning model serves multiple functions: it can generate images for any missing wavelength band, process various types of input data, and adapt to different satellite sensor configurations. This multi-functional approach replaces the need for multiple specialized physical sensors.
2Loss of information
If multiple image sensors are mounted to capture different wavelength bands, then image information completeness is improved, but manufacturing cost increases
Solution Approach 1:
Instead of manufacturing physical sensors for each wavelength band, the system uses software-based deep learning models to generate the necessary image data. This software approach is significantly less expensive than manufacturing and launching multiple specialized sensor systems.
Solution Approach 2:
The patent replaces the mechanical/optical sensor system with a computational approach. Rather than physically capturing light at multiple wavelengths, the system uses algorithms to synthesize the necessary spectral information from available data, substituting mechanical hardware with software processing.
3Productivity
If conventional sensor systems are used, then image data can be obtained directly, but reliability decreases due to potential sensor failures or noise
Solution Approach 1:
The system prepares for potential sensor failures by having deep learning models that can generate replacement data. The models are trained in advance on comprehensive datasets, so when sensor failures occur, the system can immediately switch to synthetic data generation without losing productivity or reliability.
Solution Approach 2:
The deep learning models create synthetic copies of sensor data that can replace failed or noisy measurements. These generated images replicate the quality and characteristics of actual sensor data, providing reliable alternatives when physical sensors fail.
4Reliability
If deep learning-based image fusion is used, then reliability is improved by generating missing band data, but computational processing requirements increase
Solution Approach 1:
The system applies deep learning only when and where needed - specifically when sensor data is missing or corrupted. Rather than continuously processing all data through complex models, the system processes only the necessary portions, reducing overall computational energy consumption while maintaining reliability.
Data Source
AI summary
The present invention relates to a multi-band image generation system and method using a deep-learning network, and to a technology of generating a multi-band image based on an image of a panchromatic band by analyzing a nonlinear relationship between the panchromatic band and each multi-band using a deep-learning network.