Unsupervised Computer Vision for Aerospace Systems

This book addresses perception and monitoring challenges in aerospace systems by employing innovative unsupervised learning techniques, thereby providing solutions for scenarios characterized by limited labelled data or dynamic environments. It explores practical methods such as domain adaptation for cross-modal pose estimation, causal inference for point cloud segmentation, and lightweight vision models optimized for edge computing. Key features include algorithm flowcharts, performance comparison tables, and real-world case studies covering planetary crater detection and spacecraft pose estimation. The integration of generative adversarial networks (GANs) for satellite jitter estimation and multistep adaptation strategies for defect detection offers actionable insights, supported by real industrial datasets, embedded hardware schematics, software code snippets, and optimization guidelines for real-time deployment. Engineers and researchers will obtain tools to enhance robustness across modalities and domains, ensuring generalizability in resource-constrained settings. This book serves as a valuable reference for aerospace engineers, computer vision specialists, and remote sensing practitioners and also empowers aerospace infrastructure inspectors adopting advanced vision technologies.