The Use of Artificial Intelligence for Space Applications

Preface
Acknowledgements
Contents
SAR Image Formation: Conventional and AI-Based Approaches on Sentinel-1 Raw Products
1 Introduction
2 Fundamentals
2.1 Synthetic Aperture Radar
2.2 Sentinel-1 Mission
2.3 Deep Learning
3 SAR Focusing State-of-the-Art Methods
3.1 Conventional SAR Image Formation Techniques
3.2 Deep Learning for SAR
4 SAR Image Formation Through Conventional Approach
4.1 StripMap Products
4.2 Interferometric Wide Swath Products
5 SAR Image Formation Through AI-Based Approach
5.1 Models Definition and Training
5.2 Preliminary Testing Phase
6 Conclusions
References
YOLO v4 Based Algorithm for Resident Space Object Detection and Tracking
1 Introduction
2 Methodology
2.1 Algorithm Design and Configuration
2.2 Algorithm Implementation Methodology
2.3 Dataset Creation
3 Training, Testing and Results
3.1 Comparison on Real RSOs Passages
3.2 Test of a Real RSO Passage
3.3 Test with High Fidelity Star Tracker Simulator Images
4 Conclusion
References
GPU@SAT: A General-Purpose Programmable Accelerator for on Board Data Processing and Satellite Autonomy
1 Introduction
2 Hardware Architecture
3 Software Architecture
4 AI-FDIR Algorithms
5 Preliminary Results and Conclusion
References
Overview of Meta-Reinforcement Learning Methods for Autonomous Landing Guidance
1 Introduction
2 Reinforcement Learning and Meta-Reinforcement Learning
3 Image-Based Moon Landing with Meta-RL
3.1 Simulation Environment
3.2 Results
4 Adding Hazard Detection and Landing Site Selection
4.1 Hazard Detection and Landing Site Selection Algorithm
4.2 Guidance Architecture
4.3 Results
5 Conclusions and Outlook
References
Satellite IoT for TT&C and Satellite Identification
1 Introduction
2 Methodology
2.1 Reference Scenario
2.2 Used Air Interface Protocols
3 Results
4 Conclusion
References
Hardware-in-the-Loop Simulations of Future Autonomous Space Systems Aided by Artificial Intelligence
1 Introduction
2 The MONSTER Robotic Facility
3 Enabling Edge Computing with AI Accelerators
3.1 Preliminary AI Deployment Analysis and Results
4 The Lunar Landing Test Case
4.1 MONSTER Setup
4.2 The Autonomous GNC Subsystem
4.3 On-Board Edge-Computing Simulation with the Jetson TX2
4.4 Transfer Learning Experiments
5 Remote Sensing Test Cases
5.1 Wildfire Classification
5.2 Volcanic Eruption Detection
5.3 Designing Future Missions for Real-Time Extreme Events Management
6 Preliminary Comparison Between FPGA and Jetson TX2 GPU
6.1 Large Fully Connected Network
6.2 Deep Convolutional Network
7 Discussion and Conclusion
References
Deep Reinforcement Learning for Pin-Point Autonomous Lunar Landing: Trajectory Recalculation for Obstacle Avoidance
1 Introduction
2 Purpose Statement and Contributions: The Need for Precision
3 Simulator Overview
3.1 Reconstructing Terrains: Titan
3.2 The Lander
4 Methodology: Soft-Actor Critic
4.1 SAC Algorithm
5 Experiment Design
5.1 Environment Setup
5.2 Experimental Results
5.3 Comparison with Other State-of-the-Art Deep Reinforcement Learning Methods
6 Ongoing Work
7 Conclusion
References
Imbalanced Data Handling for Deep Learning-Based Autonomous Crater Detection Algorithms in Terrain Relative Navigation
1 Introduction
2 Related Works
2.1 Autonomous Crater Detection with Deep Learning
2.2 Absolute and Relative Navigation
2.3 Imbalanced Dataset Handling
3 Dataset
3.1 Imbalanced Dataset
4 Methodology
4.1 Semantic Segmentation
4.2 Segmentation Performance
5 Results
5.1 Moon DEM Training
5.2 Crater Post-Processing
6 Discussion
6.1 Data sampling Versus Loss Function Choice
6.2 BCE-Based Versus FTL-Based Training
6.3 Post-Processing Performances
7 Conclusion
References
Comparative Analysis of Reinforcement Learning Algorithms for Robust Interplanetary Trajectory Design
1 Introduction
2 Problem Statement
3 Reinforcement Learning
3.1 Proximal Policy Optimization
3.2 Twin Delayed Deep Deterministic Policy Gradient
3.3 Soft Actor-Critic
4 Numerical Results
4.1 Implementation Details
4.2 Learning Curves
4.3 Robust Trajectories
4.4 Closed-Loop Performance Analysis
5 Conclusion
References
Fault Detection Exploiting Artificial Intelligence in Satellite Systems
1 Introduction
2 Related Works
3 Methods and Materials
3.1 Marsis
3.2 Dataset Analysis and Preprocessing
3.3 System Architecture
3.4 DNN Architectures and Training
3.5 Fault Simulation
4 Results
4.1 Signal Prediction and Fault Detection
4.2 Model Complexity
4.3 Prediction Horizon
5 Conclusions
References
ISS Monocular Depth Estimation Via Vision Transformer
1 Introduction
2 Method
2.1 Architecture
2.2 Simulation Environment
3 Parameters
4 Results
5 Conclusions
References
RobDT: AI-enhanced Digital Twin for Space Exploration Robotic Assets
1 Introduction
2 Related Work
3 The ROBDT Framework
3.1 System Architecture
3.2 Simulator
3.3 Model Updaters
3.4 Planner and What-if Analysis
3.5 Fault Detection and Diagnosis
4 The ROBDT Case Study
5 Conclusions
References
An Overview of X-TFC Applications for Aerospace Optimal Control Problems
1 Introduction
2 X-TFC for Aerospace Optimal Control Problems
3 Conclusions and Outlooks
References
Innovative ML-based Methods for Automated On-board Spacecraft Anomaly Detection
1 Introduction
1.1 Related Works
2 Identified Scenario and System Requirement
3 ML Models for Anomaly Detection
3.1 Local Outlier Factor
3.2 Principal Component Analysis
3.3 One-Class Support Vector Machines
3.4 Autoencoder
3.5 Model Tuning for On-board Implementation
4 Numerical Evidences
5 Conclusion
References
Explainable AI with the Information Bottleneck Principle
1 Introduction
2 The IB Theory
3 Mutual Information and MINE
4 Experimental Setup
4.1 12bit
4.2 Resized MNIST
5 Experimental Results
6 Conclusions
References
SINAV: An ASI Study of Future AI Applications on Spatial Rovers
1 Introduction
2 Mars and Moon Challenging Environments
3 SINAV Main Requirements
4 The Value of Deep Learning
5 Terrain Traversability Approach
6 Depth Map Approach
7 Object Detection Approach
8 Opportunistic Science Approach
9 SINAV MLOps Approach
9.1 Scope Project
9.2 Define and Collect Data
9.3 Developed Models
9.4 Prepare for Production
9.5 Deploy for Production
9.6 Monitor and Feedback Loop
10 SINAV Hardware and Software Solutions
11 Metrics and Dataset Organization
12 Conclusions
References
Deep Learning for Navigation of Small Satellites About Asteroids: An Introduction to the DeepNav Project
1 Introduction
2 Framework
2.1 Optical Navigation
3 Project Development
3.1 Project Objectives
3.2 Methodology
4 Expected Outcomes
5 Conclusions
References
Object Recognition Algorithms for the Didymos Binary System
1 Introduction
2 Methodology
2.1 Datasets
2.2 Baseline IP
2.3 Convolutional Extreme Learning Machine
2.4 Convolutional Neural Network
2.5 Random Forest
2.6 Overall IP Strategy
3 Results
4 Conclusions
References
Towards an Explainable Artificial Intelligence Approach for Ships Detection from Satellite Imagery
1 Introduction
2 Data Description
3 Methodology
3.1 Convolutional Neural Network Architecture
3.2 Explainable Ship Detection Model
3.3 Performance Metrics
4 Results
5 Conclusion
References
Investigating Vision Transformers for Bridging Domain Gap in Satellite Pose Estimation
1 Introduction
2 Related Work
2.1 Monocular Pose Estimation in Space
2.2 Domain Generalization
2.3 Transformers in Vision
3 Pose Estimation Competition and SPEED+
4 Dataset Preprocessing
5 Three Stage Domain Adversarial Approach
5.1 Architecture and Training Setup
5.2 Main Experiments
5.3 Best Performances
5.4 Comparison with a CNN Based Pipeline
5.5 On Device Inference
6 Lightweight Dual Stage Domain Agnostic Approach
6.1 Model Structure
6.2 Training Setup
6.3 Effects of Data Augmentations
6.4 Comparison Between Pure Transformer Based and Hybrid Solution
6.5 On Device Inference
7 Conclusions
References
Detection of Clouds and Cloud Shadows on Sentinel-2 Data Using an Adapted Version of the Cloud-Net Model
1 Introduction
2 Related Work
3 Method
4 Experimental Setup
4.1 Dataset
4.2 Band Selection
4.3 Performance Evaluation
5 Results
6 Conclusion
References
PRISMA Hyperspectral Image Segmentation with U-Net Convolutional Neural Network Using Singular Value Decomposition for Mapping Mining Areas: Preliminary Results
1 Introduction
2 The Main Study Area
3 Pre-Processing
4 Methodology
4.1 Singular Value Decomposition
4.2 Data Augmentation
4.3 Model Selection: U-Net Convolutional Network
5 Experimental Preliminary Results
5.1 Sardinia Area Results
5.2 Brazil Area Results
6 Discussion and Conclusions
References
Earth Observation Big Data Exploitation for Water Reservoirs Continuous Monitoring: The Potential of Sentinel-2 Data and HPC
1 Introduction
2 Proposed Approach
2.1 AI Powered Super Resolution
2.2 Detection of the Horizontal Extent of the Reservoirs
2.3 Computation of the 3D Volumetric Changes
3 Testing Sites
4 Preliminary Results
5 Concluding Remarks
References
Retrieval of Marine Parameters from Hyperspectral Satellite Data and Machine Learning Methods
1 Introduction
2 Methods and Data
2.1 The Coupled Atmosphere-Ocean RTM
2.2 PRISMA and Ancillary Data
2.3 Solving the Nonlinear Inverse Problem
2.4 Machine Learning Methods
3 Results and Discussion
3.1 RTM Sensitivity to Inputs
3.2 Variational Retrieval
3.3 Machine Learning Retrieval
4 Conclusions
References
Lunar Site Preparation and Open Pit Resource Extraction Using Neuromorphic Robot Swarms
1 Introduction
2 Background
3 Artificial Neural Tissue
3.1 Motor Neurons
3.2 The Decision Neuron
3.3 Activation Function
4 Resource-Collection Task
5 Results and Discussion
5.1 Behavior Scalability
5.2 Controller Scalability
6 Conclusions
References
Artificial Intelligence for SAR Focusing
1 Introduction
2 Method
2.1 Dataset
2.2 Dataset Preparation
2.3 Normalization
2.4 Artificial Neural Networks
2.5 Loss Function, Optimizer and Hyperparameters
3 Results
3.1 Settings
3.2 Models’ Comparisons
4 Conclusions
4.1 Time and Available Resources
4.2 Future Works
References
Canopy Fire Effects Estimation Using Sentinel-2 Imagery and Deep Learning Approach. A Case Study on the Aspromonte National Park
1 Introduction
2 Study Area
3 Materials and Methods
3.1 Dataset and Pre-Processing
3.2 Field Measurements and Sampling Points Collection
3.3 Artificial Neural Network Construction and Image Classification
3.4 Accuracy Assessment
4 Results
4.1 Final Structure of the ANN
4.2 Classified Fire Effects Map
4.3 Feature Importance
4.4 Map Accuracy
5 Discussions
6 Conclusions
References
A Supervised Learning-Based Approach to Maneuver Detection Through TLE Data Mining
1 Introduction
2 Fundamentals
2.1 Two Line Element File and Orbital Parameters
2.2 Segmentation-Aimed Network Architecture
3 Method
3.1 Data Structure and Pre-Processing
3.2 Network Structure and Training
3.3 Network Post-Processing and Testing
4 Results
4.1 Network A Testing
4.2 Network B Testing
4.3 Results Comparison
5 Conclusions
References
A Machine Learning Approach for Monitoring of GNSS Signal Quality in Spaceborne Receivers: Evil Waveform and RF Threats
1 Introduction and Motivation for GNSS Space Applications
2 Observables and Working Hypothesis
2.1 Dataset Description
2.2 Live Collection and Synthetic Datasets Generation
3 Machine Learning: Support Vector Machines Scheme
3.1 The Algorithm: Cross-Validation and Performance Metrics
3.2 Considerations on Computational Complexity and Dataset Size
4 Study Case and ROC Performance
4.1 Conclusions and Way Forward
References