CubeSats provide a low-cost, convenient, and effective way of acquiring remote sensing data, and have great potential for remote sensing object detection. Although deep learning-based models have achieved excellent performance in object detection, they suffer from the problem of numerous parameters, making them difficult to deploy on CubeSats with limited memory and computational power. Existing approaches attempt to prune redundant parameters, but this inevitably causes a degradation in detection accuracy. In this paper, the novel Context-aware Dense Feature Distillation (CDFD) is proposed, guiding a small student network to integrate features extracted from multi-teacher networks to train a lightweight and superior detector for onboard remote sensing object detection. Specifically, a Contextual Feature Generation Module (CFGM) is designed to rebuild the non-local relationships between different pixels and transfer them from teacher to student, thus guiding students to extract rich contextual features to assist in remote sensing object detection. In addition, an Adaptive Dense Multi-teacher Distillation (ADMD) strategy is proposed, which performs adaptive weighted loss fusion of students with multiple well-trained teachers, guiding students to integrate the learning of helpful knowledge from multiple teachers. Extensive experiments were conducted on two large-scale remote sensing object detection datasets with various network structures; the results demonstrate that the trained lightweight network achieves auspicious performance. Our approach also shows good generality for existing state-of-the-art remote sensing object detectors. Furthermore, by experimenting on large general object datasets, we demonstrate that our approach is equally practical for general object detection distillation.

Multispectral image pairs can provide the combined information, making object detection applications more reliable and robust in the open world. To fully exploit the different modalities, we present a simple yet effective cross-modality feature fusion approach, named Cross-Modality Fusion Transformer (CFT) in this paper. Unlike prior CNNs-based works, guided by the transformer scheme, our network learns long-range dependencies and integrates global contextual information in the feature extraction stage. More importantly, by leveraging the self attention of the transformer, the network can naturally carry out simultaneous intra-modality and inter-modality fusion, and robustly capture the latent interactions between RGB and Thermal domains, thereby significantly improving the performance of multispectral object detection. Extensive experiments and ablation studies on multiple datasets demonstrate that our approach is effective and achieves state-of-the-art detection performance. Our code and models are available at https://github.com/DocF/multispectral-object-detection.

This work presents a novel self-organizing control method for mega constellations to meet the continuous Earth observation requirements. In order to decrease the TT&C pressure caused by numerous satellites, constellation satellites are not controlled according to the designed configurations but are controlled with respect to intersatellite constraints. By analyzing the street-of-coverage (SOC) of coplanar constellation satellites, the continuous coverage constraint of the mega constellation is transformed into constraints of the right ascension of ascending node (RAAN) and relative motion bound between every two adjacent coplanar satellites. The proposed continuous coverage constraint can be satisfied by most ongoing or planned mega constellations. Artificial potential functions (APFs) are used to realize self-organizing control. The scale-independent relative orbital elements (SIROEs) are innovatively presented as the self-organizing control variables. Using the Gaussian equations and Lyapunov’s theory, the stability of the APF control in quadratic form is proven, from which it can be concluded that the APF control variables of the controlled satellite should have the same time derivative as the target satellite states under two-body Keplerian motion condition, and SIROEs are ideal choices. The proposed controllers and self-organizing rules are verified in the sub-constellation of the GW-2 mega constellation by simulation. The results demonstrate the goodness in control effect and ground coverage performance.

Real-time astronaut visual tracking is the most important prerequisite for flying assistant robot to follow and assist the served astronaut in the space station. In this paper, an astronaut visual tracking algorithm which is based on deep learning and probabilistic model is proposed. Fine-tuned with feature extraction layers’ parameters being initialized by ready-made model, an improved SSD (Single Shot Multibox Detector) network was proposed for robust astronaut detection in color image. Associating the detection results with synchronized depth image measured by RGB-D camera, a probabilistic model is presented to ensure accurate and consecutive tracking of the certain served astronaut. The algorithm runs 10 fps at Jetson TX2, and it was extensively validated by several datasets which contain most instances of astronaut activities. The experimental results indicate that our proposed algorithm achieves not only robust tracking of the specified person with diverse postures or dressings but also effective occlusion detection for avoiding mistaken tracking.

In this study, a model predictive control (MPC) method is developed for a servicer spacecraft autonomously approaching a tumbling failed spacecraft at an ultraclose range. Flight safety and collision avoidance are basic requirements during the approach. Two types of a failed spacecraft with complex configurations are considered, and a double-ellipsoid composite envelope strategy is designed to model their keep-out zones. Given the keep-out zone of the servicer, two expanded ellipsoids are subsequently introduced to determine the collision and sufficient conditions for collision avoidance are derived by using the form of concave constraint. The tumbling motion of the target is considered, and a CW-based translational dynamics and derived attitude dynamics of the target are formulated to predict the motion of the docking point and keep-out zone. The MPC is formulated to drive the servicer tracking the docking point with collision avoidance and handle constraints including control input saturation and relative velocity bound. Convexification of the collision avoidance constraint and sequential convex programming are adopted for the implementation of MPC. Scenarios on the servicer with different initial positions approaching the target with different angular velocities are simulated, and the simulation results indicate that the proposed MPC method is effective.