Resiliency and robustness study of the Apache storm-based distributed resilient remote sensing platformZhao, Qi; Cheng, Cheng-Yu; Wu, Cheng-Ying; Yang, Yuchen; Qureshi, Muhammad A.; Liu, Hang; Chen, Genshen
doi: 10.1117/12.3053505pmid: N/A
In mission-critical environments, real-time data processing and resilience to system failures are essential for maintaining situational awareness and operational continuity. Our distributed resilient remote sensing platform, a robust real-time distributed data processing system, addresses these challenges by leveraging Apache Storm framework. Our platform integrates Microsoft HoloLens and distributed computing components to support enhanced situational awareness and efficient data handling in critical operational environments. While Apache Storm’s default configurations struggled with recovery during node failures and network disconnections, we addressed these limitations through the design, development, and evaluation of our distributed resilient mobile sensing platform. The paper further discusses the methods used to test and evaluate the platform’s resiliency under different failure scenarios, demonstrating the system’s rapid recovery and sustained functionality, even under significant stress. Additionally, we explore parameter tuning techniques aimed at improving the platform’s recovery speed, specifically in scenarios involving node failures and network disconnection. The results show a significant reduction in recovery time, validating the platform’s robustness and scalability in real-world applications. These findings underscore the potential of our platform for military and emergency response operations, where continuous, reliable data processing is essential.
Design and experimental validation of a high-altitude 2-DoF robotic tracking system for the air-LUSI missionMcCafferty-Leroux, Alex; Newton, Andrew; Maxwell, Stephen; Gadsden, S. Andrew; Turpie, Kevin R.
doi: 10.1117/12.3053846pmid: N/A
Earth observing satellites are responsible for a variety of essential operations, including communication, navigation, and scientific observation. In the harsh environment of space, however, on-orbit sensor degradation can cause inaccuracies over time. Using the Moon as a reference source, we can achieve reliable in-situ satellite calibration to address this. The objective of the airborne lunar spectral irradiance (air-LUSI) mission is to improve the value of the Moon in this regard, extending the accuracy of existing lunar models and observations through its establishment of an in-situ, SI-traceable, absolute calibration target. A non-imaging telescope is flown at 70,000 ft aboard an ER-2 aircraft, circumventing a significant fraction of lunar spectra measurement uncertainties due to Earth’s atmosphere. The improvement in these measurements is projected to improve the widely accepted ROLO (GIRO) model such that it achieves absolute uncertainties of less than 1%. Due to erratic motion of the aircraft, a robotic telescope mount is deployed to stabilize the telescope along its line-of-sight, minimizing data loss. This paper discusses the development and in-lab performance of the tracking system for this instrument, referred to as the high-altitude aircraft-mounted robotic telescope mount (HAAMR). An overview of the robotic system is provided, and a static loading analysis is conducted to demonstrate airworthiness. This instrument was designed to be an improvement upon the ARTEMIS subsystem of the same purpose, for the series of flight campaigns from December 2024 to December 2025. The analysis demonstrates, from past flight data, that the tracking accuracy of the HAAMR will guarantee high pointing accuracy during turbulent flight.
Multi-objective system design of an active imaging mode for a sparse aperture telescopeBadura, Gregory P.; Hope, Douglas; Plis, Elena; Leger, James; Prasad, Sudhakar; Kuhn, Jeff; Arunkumar, Ebenezer; Jefferies, Stuart
doi: 10.1117/12.3052291pmid: N/A
Imaging of small and distant targets under conditions where limited photons passively reach the target can be extremely challenging due to resolution and signal-to-noise ratio constraints. Conventional small aperture telescopes have an inability to spatially resolve these types of targets according to the Rayleigh criterion. Large aperture telescopes, on the other hand, can resolve such targets but are prohibitively expensive to produce. An alternative solution known as optical interferometry can potentially allow for high resolution imaging without the extreme cost of large aperture systems via its use of a sparse aperture array. A sparse aperture imaging array enhances resolution while minimizing the total light collection area by combining the light collection capabilities of distributed small apertures. While the sparse aperture array can potentially solve the resolution problem, it does not solve the problem of limited photons being reflected by the target. In this study, we therefore perform an initial exploration on the feasibility of combining a sparse aperture array with active imaging techniques to achieve imaging of targets with small angular extents under photon-starved conditions. A Multi-Objective Optimization (MOO) framework is developed where the sparsity of the sub-apertures, the wavelength of the actively propagated pulse, and the energy of the actively propagated pulse are optimized under the consideration of target signal-to-noise, beam combining difficulty, and imaging resolution constraints. We discuss the implications of this initial design on factors such the Point Spread Function (PSF) angular extent, target contrast, and expected photons returned to the telescope aperture after round-trip beam propagation.
Advanced path planning and collision avoidance for quadcopter drones using deep Q-learning in 3D simulationJackson, Carlan; Fu, Yujian; Khan, Simon
doi: 10.1117/12.3065450pmid: N/A
This paper presents the development and implementation of a drone system designed for efficient path planning and collision avoidance. Utilizing Unreal Engine to create a simulation environment and Deep Q-Learning (DQN) for decision-making, we detail our approach to building a robust drone navigation system. The methodology includes developing a 3D environmental design, drone and environment configuration, algorithm development for collision avoidance, and optimization of DQN training parameters. Through a comprehensive evaluation, we were able to demonstrate promising improvements in the drone’s ability to navigate complex environments while effectively avoiding obstacles. Performance evaluation showed a steady increase in reward values during training. Cumulative rewards rose from 1.92 to 8.18, with a maximum possible reward of 10 per episode. This indicates successful navigation toward the goal coordinate as well as it reflects improvement over time as the drone learned from its environment. As training concluded and parameters were tuned, testing metrics validated the drone’s ability. The majority of testing episodes yielded the maximum reward of 10. This indicates that the drone successfully reached its goal. As a result, these findings provide a strong foundation for our future research and advancements in autonomous drone systems.
Improved Kalman filtering through moment-based innovation gain strategiesHilal, Waleed; McCafferty-Leroux, Alex; Gadsden, Stephen A.; Yawney, John
doi: 10.1117/12.3053779pmid: N/A
This paper presents the moment-based Kalman filter (MKF), a novel sub-optimal estimation strategy designed to enhance robustness in systems subject to modeling uncertainties or external disturbances. Unlike the conventional Kalman filter, the MKF incorporates higher-order statistical moments of the innovation to inform its gain calculation, allowing for a more nuanced representation of the underlying noise and measurement error characteristics. The filter is structured as a predictor-corrector algorithm and maintains computational efficiency while offering improved adaptability in uncertain environments. A mathematical formulation of the MKF is provided, along with a proof of stability. Performance is evaluated using a simulated electrohydrostatic actuator (EHA) model undergoing a leakage fault. Results from the computational study demonstrate that the MKF provides more accurate state estimates than the standard Kalman filter, particularly under faulty or uncertain operating conditions.
Optimizing satellite constellations for rapid revisit interferometric synthetic aperture radar (InSAR)Weerts, Grant; Gustafson, Michael; Smith, David; Boyarsky, Michael
doi: 10.1117/12.3051839pmid: N/A
Spaceborne Interferometric Synthetic Aperture Radar (InSAR) is a critical technique for a range of Earth observation (EO) applications, including high-resolution topographical mapping, disaster monitoring, and global environmental change detection. Demand for InSAR has led to a need for techniques that can optimize the orbits of a constellation of satellites to create frequent revisit opportunities for interferometric measurements. Planning the orbits for a small number of satellites dedicated solely to InSAR is straightforward, but doing so can negatively impact general-use revisit and coverage. Ideally, a SAR constellation can do both well, navigating the tradeoff between general coverage needs and frequent InSAR revisits. In this work, we explore the impact of spatial arrangement and orbital parameters on satellite constellation performance. We consider a variety of constellation types, including Walker-Delta, single-plane, and multiplane formations, to identify configurations that are best suited to specific mission objectives. By analyzing and generalizing the orbital characteristics and formations that define high-quality constellations, we reach conclusions on how to best plan orbits for specific imaging goals. We further consider practical constraints of implementing such constellations, including launch altitudes and inclinations available from commercial satellite launches. These findings are intended to inform the planning of future satellite missions, supporting a wide range of high-impact global applications, especially as rapid environmental5 and hazard monitoring becomes increasingly critical.
Front Matter: Volume 13483doi: 10.1117/12.3072424pmid: N/A
This PDF file contains the front matter associated with SPIE Proceedings Volume 13483, including the Title Page, Copyright information, Table of Contents, and Conference Committee information.
Machine learning (ML) prediction of cislunar family and period on non-uniformly sampled line of sight time seriesBadura, Gregory P.; Arunkumar, Ebenezer; Velez-Reyes, Miguel; Ho, Koki
doi: 10.1117/12.3052292pmid: N/A
Angles-only Initial Orbit Determination (IOD) techniques developed for near-Earth objects do not readily extend to the cislunar domain due to their reliance on two-body gravitational assumptions. For objects that are beyond Geostationary Orbit (GEO), the Moon’s gravitational influence becomes significant and requires three-body gravitational models. Under three-body dynamics, many near-Earth IOD assumptions fail: orbits are not necessarily periodic, planar, or elliptical. Emerging Machine Learning (ML) algorithms such as Physics Informed Neural Networks (PINNs) have recently demonstrated promise for accomplishing cislunar IOD using line of sight measurements. PINNs perform supervised prediction of cislunar satellite position while simultaneously respecting laws of orbital dynamics that are approximated using automatic differentiation algorithms. Unfortunately, recent research has shown that PINNs are sensitive to the initial trajectory estimate. For example, research has shown that if a PINN’s initial trajectory estimate is in the wrong pseudopotential zone, the PINN can fail to converge due to the network’s dynamics loss spiking as it’s trajectory estimate traverses the cislunar volume. In order to improve the convergence rate of PINNs for performing IOD, the weights of the NN component must be initialized such that the initial trajectory estimate exhibits similarity in terms of the cislunar family and period to the true trajectory. As a means of aiding ML-based IOD methods, we therefore test two architectures for classification of cislunar family and intra-family period from irregularly sampled line of sight measurements. The first is a Residual Network (ResNet)-based architecture, which uses several residual blocks of convolutional filters in order to extract a feature embedding. The second is a time-variant encoder architecture that relies on Recurrent Neural Networks (RNNs) to extract a time-flattened feature embedding. Our results indicate that both architectures performed well at predicting cislunar family. The time-variant encoder model and ResNet models achieved 94% and 95% accuracy, respectively, in cislunar family classification depending on the number of line of sight measurements (Nt) available to the classifier. Our results also show that both architectures can accomplish the goal of intra-family period estimation. The time-variant encoder model and ResNet model both achieved ≤ 1% error in average period estimation across cislunar families as the number of line of sight measurements increased to a maximum of Nt = 36. By merging these classification systems with a PINN for IOD, we demonstrate that a repeatable and accurate end-to-end ML system for performing cislunar OD can be attained.
An investigation of PyTorch image models for remote sensing image classificationChen, Hua-mei; Fan, Zhengyang; Blasch, Erik; Pham, Khanh; Khan, Simon; Chen, Genshe
doi: 10.1117/12.3053506pmid: N/A
The PyTorch IMage Models (TIMM) library, developed by Ross Wightman, serves as a foundational deep learning toolkit that encompasses a diverse range of state-of-the-art computer vision models, layers, utilities, optimizers, schedulers, data loaders, augmentations, and training/validation scripts conducive to reproducing ImageNet training results. As of June 2024, TIMM had boasted over 1,000 pre-trained models, including Convolutional Neural Networks (CNNs), Vision Transformers (ViTs), and hybrid architectures of CNNs and ViTs. In this study, we evaluated 1,365 TIMM pre-trained models for the task of classification on remote sensing images. To assess TIMM’s performance, the K-Means clustering accuracy was calculated for each model, documenting both accuracy and model size. The analysis uses the full UC Merced Land Use (UCM) dataset (21 classes) and the Aerial Image Dataset (AID) (30 classes) for testing. Given the variation in output feature dimensions across models, we applied principal component analysis (PCA) to reduce the dimensionality to 128 before calculating clustering accuracy. Additionally, since K-Means clustering is stochastic, we repeated the process ten times and reported the best result. Our findings indicate that: 1) the highest-performing models are predominantly ViT-based; 2) while larger models generally yield better results, several smaller models also perform competitively; and 3) there is a strong correlation between performance on the UCM and AID datasets, as shown by a Pearson correlation coefficient of ~0.77 with a p-value of less than 0.00001. These results suggest that the pre-trained models from TIMM are broadly effective for classifying remote sensing images.
Derivative-free model reference adaptive control of an uncertain satellite system based on RBF neural networksMcCafferty-Leroux, Alex; Wu, Yuandi; Kosierb, Patrick; Gadsden, S. Andrew
doi: 10.1117/12.3053855pmid: N/A
The capability for orbital satellite systems to compensate for faults and disturbances in uncertain environments is critical. Considering their applications in communication, defense, and scientific observation, the desire to reduce the negative effects due to disturbances and uncertainties has led to the development and implementation of adaptive attitude control theory. Applying this field of control, accuracy, stability, and desired performance is achievable in imperfect situations. This paper proposes the novel application of the derivative-free variation of model reference adaptive control (MRAC) to nonlinear satellite systems for trajectory tracking scenarios. Using a radial basis function (RBF) based neural network to parameterize uncertainties, this method implements a time-varying weight matrix and derivative-free update law, expected to achieve fast adaptation. This method is compared to other MRAC strategies under the same system and environment. The experimental results demonstrate the adaptability of this scheme for satellite systems, able to track trajectories with high accuracy under the presence of common environmental variables and modeling errors.