Reinforcement learning based dynamic distributed routing scheme for mega LEO satellite networks

Recently, mega Low Earth Orbit (LEO) Satellite Network (LSN) systems have gained more and more attention due to low latency, broadband communications and global coverage for ground users. One of the primary challenges for LSN systems with inter-satellite links is the routing strategy calculation and maintenance, due to LSN constellation scale and dynamic network topology feature. In order to seek an efficient routing strategy, a Q-learning-based dynamic distributed Routing scheme for LSNs (QRLSN) is proposed in this paper. To achieve low end-to-end delay and low network traffic overhead load in LSNs, QRLSN adopts a multi-objective optimization method to find the optimal next hop for forwarding data packets. Experimental results demonstrate that the proposed scheme can effectively discover the initial routing strategy and provide long-term Quality of Service (QoS) optimization during the routing maintenance process. In addition, comparison results demonstrate that QRLSN is superior to the virtual-topology-based shortest path routing algorithm.     

A policy iteration method for improving robot assembly trajectory efficiency

Bolt assembly by robots is a vital and difficult task for replacing astronauts in extra-vehicular activities (EVA), but the trajectory efficiency still needs to be improved during the wrench insertion into hex hole of bolt. In this paper, a policy iteration method based on reinforcement learning (RL) is proposed, by which the problem of trajectory efficiency improvement is constructed as an issue of RL-based objective optimization. Firstly, the projection relation between raw data and state-action space is established, and then a policy iteration initialization method is designed based on the projection to provide the initialization policy for iteration. Policy iteration based on the protective policy is applied to continuously evaluating and optimizing the action-value function of all state-action pairs till the convergence is obtained. To verify the feasibility and effectiveness of the proposed method, a noncontact demonstration experiment with human supervision is performed. Experimental results show that the initialization policy and the generated policy can be obtained by the policy iteration method in a limited number of demonstrations. A comparison between the experiments with two different assembly tolerances shows that the convergent generated policy possesses higher trajectory efficiency than the conservative one. In addition, this method can ensure safety during the training process and improve utilization efficiency of demonstration data.     

Pose matters: Pose guided graph attention network for person re-identification

Person re-Identification (reID), aiming at retrieving a person across different cameras, has been playing a more and more important role in the construction of smart city and social security. For deep-learning-based reID methods, it has been proved that using local feature together with global feature could help to give robust representation for person retrieval. Human pose information can provide the locations of human skeleton to effectively guide the network to pay more attention to these key areas, and can also help to reduce the noise distractions from background or occlusions. Based on human pose, a Pose Guided Graph Attention (PGGA) network is proposed in this paper, which is a multi-branch architecture consisting of one branch for global feature and two branches for local key-point features. A graph attention convolution layer is carefully designed to re-assign the contribution weight of each extracted local feature by modeling the similarity relations. The experimental results demonstrate the effectiveness of our approach on discriminative feature learning. Our model achieves the state-of-the-art performance on several mainstream evaluation datasets. A plenty of ablation studies and different kinds of comparison experiments are conducted to prove the effectiveness of this work, including the tests on occluded datasets and cross-domain datasets. Moreover, we further design supplementary tests in practical scenario to indicate the advantage of our work in real-word applications.     

深空极低温服役环境下材料力学/物理性能低成本快速评价方法研究进展

随着我国空间探测卫星、深空探测飞行器等国家战略装备的快速发展,宇航材料的服役环境越来越苛刻,常常面临极端低温的考验。然而低温下材料的力学/物理性能和常温相比有很大差异,极低温使役环境下材料力学/物理性能评价成为人们关注的重点问题。本文着重对基于力-热能量密度等效原理而发展的深空极低温服役环境下材料屈服强度、弹性模量、理想拉伸强度、硬度等力学性能的低成本快速评价方法研究进展进行了总结和回顾;同时,介绍了力-热能量密度等效原理在低温下半导体材料的带隙能、折射率、拉曼频移,金属材料磁晶各向异性常数等物理性能的低成本快速评价方面的推广应用情况。该评价方法为实现地面实时监测在轨航天器中关键材料的主要力学/物理性能提供了有效途径,为国家高端技术装备建设中关键材料的设计、可靠性评价和实时性能监测等提供重要的手段。     

Time resolved study of laser triggered electric discharge spark in atmosphere: Machine learning approach

In this paper we analyze the possibilities of using machine learning algorithms for analysis of optical spectra of electric discharge spark in atmosphere. Breakdown in air can be initiated by intense laser pulse, making plasma which has a significant electrical conductivity. The formed plasma can be further maintained by electric current obtained from capacitor discharge. In such a case the capacitor voltage can be much lower than the striking voltage (the voltage needed to initiate the electric breakdown in air). Present setup has timing precision and low jitter of fast laser and arbitrary high energies corresponding to capacitance and voltage to which the capacitor is charged. We have used a streak camera equipped with a spectrograph to analyze optical emission of plasma obtained in this way. Q-switched Nd:Yag laser was used to achieve the initial breakdown in air. Machine learning methods were used in order to classify optical spectra of plasmas with different electron temperatures obtained with different excitation energies. We have shown that, instead of using the usual way of identifying the spectral peaks and calculating their intensity ratio, it is possible to train the computer software to recognize the spectra corresponding to different electron temperatures. Principal component analysis was used to reduce the dimensionality of problem. We present possibilities of plasma electron temperature estimation based on several clustering algorithms.     

遥感场景分类的批处理协方差神经网络方法

针对卫星遥感图像场景分类数据集中存在的局部区域特征异常问题，提出一种采用批处理协方差层的神经网络(CovNN)模型进行遥感场景分类的方法。该方法通过计算全输入通道的局部区域均值实现一种3D批处理协方差算法，能够有效消除局部区域均值的影响，从而更好地处理局部光照过强和局部区域存在无关特征的问题。将其应用于存在局部光照异常和局部无关特征问题的卫星采集AID数据集和NWPU RESISC45数据集中，实验表明CovNN在两个数据集上均取得了超过现有卷积神经网络(CNN)的召回率，可有效降低图像局部区域特征异常的不利影响。     

基于机器学习的二次电子发射唯象模型

基于机器学习和深度人工神经网络（artificial neural network，ANN）提出一种二次电子发射唯象模型。利用Vaughan模型生成先验数据集，用于训练生成描述二次电子发射一般规律的先验知识ANN模型，并在不同参数条件下验证了先验知识ANN模型的正确性。然后，分别利用银和铝合金材料的二次电子发射系数实验数据修正先验知识ANN模型，分别得到了描述两种材料的特异ANN模型。测试结果表明，特异ANN模型计算结果与实验结果相比的平均绝对误差较Vaughan模型和Furman模型降低了30%以上，与复合唯象模型精度相当或更高。在小样本条件下测试了二次电子发射ANN模型的正确性，验证了分步训练方式的有效性和二次电子发射ANN模型对于小样本集的适应性。提出的基于机器学习的二次电子发射唯象模型能够避免复杂的参数修正过程，能够基于先验知识提升模型对于小样本的适应性，能够实现二次电子发射系数的连续插值，适于在数值模拟软件中使用。     

基于注意力机制特征重建网络的舰船目标检测

深度学习为遥感领域诸多应用提供了重要的技术支撑,光学遥感图像的舰船目标检测对国防侦察和预警具有重要意义。真实场景中的舰船往往呈不同方向任意排列,且小目标的占比大,经典的深度学习目标检测算法在这种复杂条件下精度低、易漏检。为此,本文设计了基于注意力机制特征重建网络的舰船目标检测算法。首先,通过引入注意力机制对多尺度特征融合网络模型进行训练,以高召回率产生水平锚框;然后,旋转锚框以缓解密集排列目标引起的噪声问题,并利用特征重建模块来缓解特征不对齐的问题,实现模型精炼。在HRSC2016和DOTA数据集上的测试结果表明:舰船目标检测平均精度分别达到90.20和87.52,相比经典的深度学习目标检测算法得到了有效提升,并在模拟星载嵌入式智能图像处理平台上验证了算法在轨应用的可行性。