Deep learning methods have achieved great success in tasks like image classification, object detection and fault diagnosis. However, practical limitations often prevent gathering large amounts of data. Hence, there is a recent focus on algorithms for learning with small samples. This review aims to explain popular small sample learning methods and how they perform in real world applications. The review covers different approaches like metric models, memory models, parameter updating models and sample augmentation models, discussing their pros and cons. It also explores how these methods are applied in tasks like image classification, object detection, semantic segmentation and fault diagnosis. Lastly, it discusses the limitations of small sample learning methods and predicts future research trends focusing on less data dependency, more efficient algorithms and robust models.

The strictly regressive orbit is designed under high precise orbit dynamic with high order non spherical gravitational field of earth. The strictly regressive orbit is one kind of sun synchronous repeat track orbits, of which the beginning and ending space trajectories over repeat period in WGS84 coordinates are highly coincided. The on borne tube navigation algorithm sets a strictly regressive orbit designed before launch as reference orbit, based on which a virtual formation is formed with the real satellite. Since the effect of air drag is not considered in the design process of strictly regressive orbit, the satellite in orbit can calculate the orbit decay situation and orbit decay rate caused by air drag. To reuse the reference orbit, the effects of third body perturbation of sun and moon are also not considered in the design process of strictly regressive orbit. Therefore the long period motion characteristics of orbit inclination are determined during in orbit flying. Based on the telemetering data of LT1 satellite, the orbital characteristics mentioned above are analyzed in this paper.

The Ju Mang Satellite is mainly used for carbon monitoring of terrestrial ecosystems, investigation and monitoring of terrestrial ecology and resources, and monitoring and evaluation of national major ecological projects. The Ju Mang control subsystem adopts a high precision attitude determination, high stable attitude control and hybrid trajectory planning attitude maneuver algorithm. According to the demand of the load for the monthly calibration, an inertial scanning mode is designed, which can pass through any space position at any time. According to the requirements of satellite autonomous mission planning, a surface attribute prediction method based on the surface model map is designed. For the first fire detection sensor, a high precision calculation of the geographical longitude and latitude of the fire point is designed, and relative surface linear velocity and the auxiliary cloud judgment information such as the solar zenith angle, satellite zenith angle and relative azimuth angle are provided for the sensor to dynamic fire point prediction. In orbit, the highly reliable fire pointing calculation and false fire point are realized. Based on on orbit data of Ju Mang satellite, the implementation of specific indexes is proposed. Furthermore, the scheme of control system is tested on orbit.

For the fast analysis of dynamic stray light of the flexible solar array vibration measurement camera in the space station, the relative angles between three environmental light sources(the sun, the moon and the earth) and optical axis of the camera in the orbital period are computed, meanwhile the dynamic vibration displacements of a large flexible array are simulated, a dynamic stray light discrimination algorithm that significantly reduces the amount of computation is given, based on the analysis of environmental light sources characteristics (point source and area source) and two types of stray light interference ways (direct incidence and incidence after primary reflection). The problem of stray light trajectory changing dynamically with the solar array is solved. The simulation results show that the realtime and accuracy of this algorithm can meet the stray light analysis requirements of large flexible solar array vibration measurement cameras.

The inertial measurement unit (IMU) deployed on the reusable spacecraft not only meets the needs of long term space segment, but also needs to have high dynamic performance to meet the needs of high precision navigation during reentry. According to the needs of IMU deployed on the reusable spacecraft， the methods of dynamic accuracy design and verification in the development of reusable spacecraft IMU are summarized, and a six degree of freedom vibration IMU accuracy test method is introduced. The method can simulate the reentry mechanical environment more realistically, and can be triggered by system requirements to comprehensively evaluate the dynamic accuracy of IMU. It can also provide reference for other types of reusable spacecraft.

In order to construct a longer term constraint in the process of visual positioning, and facilitate the establishment of globally consistent trajectory estimation for aircraft equipped with visual positioning equipment, a lightweight weakly supervised simultaneous localization and mapping (SLAM) loop closure detection algorithm is proposed based on contrastive learning. By establishing consistent global descriptors for images and using similarity measures to judge loops in trajectories, long term timing constraints are established. Considering the application on resource constrained platforms, an EfficientNet (efficient neural network) is used to achieve more efficient feature extraction, and the NSE (need squeeze and excitation) attention module is combined to improve the screening of effective data in the process of data dimensionality reduction. Global descriptor is integrated through a complete VLAD (vector of locally aggregated descriptors) layer. The network model can still have efficient recognition ability in environmental changes such as lighting conditions, viewing angles, and seasons. The experimental results show that, while maintaining a 2% difference in the TOP 5 recall index compared to the baseline model, the proposed method can effectively reduce model volume by 57%, training time by 35%, and improve execution efficiency by 48%, which is beneficial for deployment on resource constrained embedded platforms such as small drones

In order to address the path planning problem of intelligent agents in complex environments, this paper proposes an online off policy deep reinforcement learning algorithm model based on an improved prioritized experience replay method. Firstly, the model utilizes a flexible action evaluation algorithm to achieve collision free path planning for the intelligent agent by designing the state space, action space, and reward function. Secondly, by calculating the sample mixing priority using the sample priority and TD error, a measure of sample diversity is obtained, and an improved prioritized experience replay method based on the flexible action evaluation algorithm is proposed to enhance the learning efficiency of the model. The simulation experimental results validate the effectiveness of the proposed improved flexible action evaluation algorithm under various parameter combinations and the superiority of the improved prioritized experience replay method in model learning efficiency for continuous control tasks

In the problem of two spacecraft rendezvous, the multi impulse strategy can often achieve less fuel consumption. In this paper, the artificial bee colony algorithm is used to optimize the fixed time multi impulse rendezvous of spacecraft, and an improved artificial bee colony algorithm is proposed. This method can simplify the processing of time constraints and obtain solutions that meet the constraints of earth radius. This method is easy to program and robust, and can be applied to different perturbation models. Simulation results show that the algorithm is not easy to fall into local optima, and has higher accuracy than particle swarm optimization algorithm and traditional artificial bee colony algorithm.

Aiming at the problem of mechanical resonance caused by weak damping flexible mode of sandwich system and the influence of backlash nonlinear factors, an auto coupling PID control method independent of the controlled object is used in this paper. This method treats all uncertain factors such as nonlinear dead zone in the sandwich system as a total disturbance, establishes a controlled error system with the total disturbance as the excitation, and then designs an auto coupled PID control system based on speed factor, and fully proves the robust stability and antidisturbance robustness of the autocoupled PID control system in theory. The simulation results show the effectiveness of the autocoupling PID control effect. The control method not only has a good response speed, but also has good robustness. Meanwhile, the method has a good application prospect in the field of sandwich structure flexible transmission system control.

Timely and effective detection of abnormal changes in spacecraft power subsystem (SPS) is an important guarantee for the safe and stable operation of spacecraft. However, due to the complex working environment of SPS and the closed loop structure inside SPS, the telemetry signal characterizing its working state contains noise and cannot reflect the fault information in time. Therefore, in view of the noise and label free problems of SPS telemetry signals, a SPS fault detection method is proposed based on multiresolution singular value decomposition (MSVD) and auto encoder (AE). Firstly, MSVD is applied to the field of wave signal denoising to reduce the influence of noise on telemetry signal. Secondly, aiming at the problem of lack of fault labels in SPS telemetry data, an unsupervised auto encoder algorithm is used to detect the abnormal data after noise reduction. Finally, the proposed algorithm is applied to SPS, and the denoising effects of MSVD, wavelet transform and empirical mode analysis are compared. Then, the SPS is detected by AE. The results show that the proposed algorithm has lower misjudgment rate and higher detection rate.

Target detection related technology has been widely used in space target surveillance, satellite automatic orbit finding and other fields. It is also one of the most important and challenging research branches in the field of computer vision and has gradually become a hot research area in the military field at home and abroad. In the modern air space confrontation, the acquisition of near earth vehicle targets through satellite remote sensing images can quickly judge the effective strength of the enemy forces, which enables our troops to occupy a strategic advantage. Aiming at the problems such as too long satellite observation distance and complex background of remote sensing image, the small sample target detection algorithm is studied based on one stage light weighted network YOLOv8. The research work of this paper mainly includes three aspects. Firstly, the generalization performance of the model is improved by image enhancement methods such as image flipping, Mosaic data enhancement and mixup data enhancement. Secondly, the average accuracy of the model is improved by adjusting the optimization function, reducing the class loss gain and reducing the mask ratio. Thirdly, the computational efficiency of the model is improved via the preset parameter and loading the model derived from the original optimization function. The method proposed in this paper is verified on the public aircraft data set, and the verification indexes include precise recall (PR), average accuracy (mAP) and the number of frames per second (FPS). The results show that the improved network model proposed in this paper can meet the needs of fast target detection in satellite remote sensing images.

With the increase of spacecraft software quantity and complexity, the demand for software reuse is becoming more and more urgent. In this paper, for the electric propulsion software of multiple platforms at the present stage, a structured analysis method is used to carry out software requirement analysis, extract common requirements and variable requirements, and identify reusable software components. After that, an event driven model based on adaptation improvement is proposed to assemble software components, and an electric propulsion software architecture framework is established. Finally, the public service operation，the form of encapsulation of hardware drivers are described, and the way that the software architecture can be applied in specific projects is given. The electric propulsion architecture framework proposed in this paper and the designed and implemented software components have been applied in several projects and achieved good results.