Neural Network Object Range Detection for Small Satellites
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object range detection techniques require costly and large hardware, which is a concern for spacecraft with limited size, weight, and power capacity, and existing software-based methods are slow to converge to a usable solution.
Innovation Solution
Implementing a neural network-based system that uses machine learning and deep neural networks to estimate object range without additional hardware, utilizing simulated training data to refine range predictions and accelerate solution convergence.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing hardware-based object range detection techniques are used, then measurement precision is improved, but device complexity and weight increase
Solution Approach 1:
The patent replaces hardware-based range detection systems with a software-based machine learning approach. A neural network model is trained to estimate object range from image data alone, eliminating the need for complex hardware systems like LIDAR or radar while maintaining acceptable measurement precision. The model processes standard camera images and outputs range estimates without requiring additional sensing hardware.
Solution Approach 2:
The patent creates a virtual model (neural network) that copies the functionality of expensive hardware range detection systems. The trained model learns to replicate the range detection capability of hardware systems through software, allowing the observation system to function as if it had the expensive hardware without actually installing it.
2Device complexity
If existing software-based object range techniques are used, then device complexity is reduced, but productivity deteriorates due to slow convergence
Solution Approach 1:
The patent performs preliminary training of the neural network model using synthetic training data generated from orbital mechanics simulations. This pre-training phase allows the model to learn range estimation patterns before deployment, so that during actual operation, the model can provide rapid range estimates without requiring slow iterative convergence during real-time observation.
Solution Approach 2:
The patent changes the operational parameters of the system by using machine learning inference instead of iterative orbital determination algorithms. This parameter change transforms the computation from an iterative process that requires multiple passes to a direct inference process that provides immediate results, dramatically improving productivity.
3Measurement precision
If additional hardware is added to improve range detection, then measurement precision is improved, but weight increases
Solution Approach 1:
The patent substitutes physical hardware components with a computational model. Instead of adding LIDAR, radar, or other active sensing hardware that would increase spacecraft weight, the system uses a neural network that processes data from existing sensors to derive range information, maintaining measurement precision without additional weight.
Solution Approach 2:
The patent makes the existing observation system multi-functional by enabling it to perform both imaging and range estimation using the same hardware components. The neural network allows the standard camera system to serve dual purposes: capturing images and estimating object range, eliminating the need for dedicated range-finding hardware.
Data Source
AI summary
A computer-implemented method may include: storing, by the computing device, information linking a dataset associated with a simulated image with an object range truth; receiving, by the computing device, an operational image from an observation system, wherein the operational image comprises the object; determining, by the computing device, a range of the object from the operational image based on the simulated image and the object range truth; and executing, by the computing device, a computer-based instruction based on the range of the object.