Satellite Constellation Distributed Remote Sensing Interpretation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The processing capability of a single satellite for remote sensing data is limited, resulting in low accuracy for interpreting ground observation data, and the challenge is to improve interpretation accuracy using multiple satellites with complementary remote sensing data from the same observation region.
Innovation Solution
A constellation-based remote sensing distributed collaborative determination method and apparatus, where multiple satellites use trained neural networks to process and fuse remote sensing data from different imaging modes, integrating feature extraction and prediction capabilities to enhance interpretation accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single satellite processes remote sensing data independently, then the processing capability is limited, but the interpretation accuracy of ground observation data is low
Solution Approach 1:
The patent merges the processing capabilities of multiple satellites by establishing a constellation-based collaborative system. Each satellite processes its own remote sensing data independently through local neural networks, then shares results with other satellites. This combining approach allows the system to achieve higher interpretation accuracy than a single satellite could achieve alone, while maintaining distributed processing capability.
Solution Approach 2:
The patent implements multi-functionality by enabling each satellite to perform both independent processing and collaborative processing. The neural network model is designed to handle multiple functions: local data processing, feature extraction, and collaborative inference. This universal design allows the system to leverage both the individual capabilities of each satellite and the collective power of the constellation.
2Measurement precision
If multiple satellites process remote sensing data independently, then the interpretation accuracy improves, but the data fusion complexity increases
Solution Approach 1:
The patent segments the data fusion process into distinct stages: each satellite first processes its own data independently through local neural networks to extract features and generate predictions, then these results are fused at the constellation level. This segmentation reduces the complexity of data fusion by breaking down the overall process into manageable independent processing units followed by systematic integration.
Solution Approach 2:
The patent introduces an intermediary mechanism in the form of a shared neural network model that serves as a mediator between individual satellite processing and final data fusion. This intermediary model provides a standardized framework for integrating data from multiple satellites, reducing the complexity of direct multi-source fusion by establishing a common processing language and structure.
3Measurement precision
If neural network models are trained separately for each satellite, then the processing speed is fast, but the model accuracy is limited
Solution Approach 1:
The patent applies preliminary action by pre-training a shared neural network model using data from multiple satellites before deployment. This preliminary training phase establishes a robust foundation model that captures general patterns across different satellites and imaging modes. During actual operation, each satellite can then use this pre-trained model for rapid local processing, achieving both high accuracy and fast processing speeds.
Solution Approach 2:
The patent utilizes parameter changes by adapting the neural network model parameters through transfer learning. The shared model parameters are updated based on each satellite's specific data characteristics while maintaining the overall model structure. This allows the system to achieve high accuracy for each satellite without requiring complete retraining, thus reducing training time while improving model accuracy.
Data Source
Figure 1~2
Figure 3
Figure 4~5
AI summary
The present disclosure relates to the technical field of data identification, and in particular to a constellation-based remote sensing distributed collaborative determination method and apparatus, and a medium and a satellite. The constellation includes n (n≥2) satellites, and the constellation includes a first satellite and a second satellite. The method includes: obtaining, by the first satellite using a trained first neural network, a first prediction result and a first feature of first remote sensing data of an observation region; and transmitting, by the first satellite, the first prediction result and the first feature to the second satellite. The second satellite performs determination based on the first prediction result, the first feature, a second prediction result and a second feature. According to the present disclosure, the interpretation accuracy can be improved based on multiple pieces of remote sensing data obtained in different imaging modes for a same observation region.