基于LSTM的飞行数据挖掘模型构建方法

提出了一种基于LSTM （Long Short Time Memory）模型的飞行历史数据挖掘模型的构建方法，此模型可以将飞行数据中有价值的目标数据自动提取出来。首先，通过滑动窗口法获得待检测数据；然后，将预先做好的训练样本数据输入到所构造的LSTM模型中进行训练，得到数据挖掘模型；最后，将待检测数据导入到训练好的LSTM模型中进行模式识别，将目标数据片段挖掘出来。结果表明，基于LSTM模型的飞行数据挖掘模型构建方法通用化程度高，可用于挖掘不同类型的目标数据，且识别率高，具有很高的工程应用价值。     

低轨卫星缺失时序数据的模式识别方法

在航天应用中，低轨卫星经常会由于原始数据缺失而影响卫星时序数据模式识别结果，降低准确率。针对该问题提出了一种新型MR-GRU模型，可有效处理缺失时序数据，并获得较好的模式识别准确率。区别于传统模型的补全缺失数据的方法，MR-GRU模型直接在缺失时序数据上运用循环神经网络进行训练，对传统门控循环单元结构进行了改进，增加了两个新变量：掩蔽项和衰减项。掩蔽项作用于输入，衰减项作用于输入和隐层单元输出。MR-GRU模型不仅能够保持时序数据固有的时间特性，还能有效提高模式识别精度。在卫星时序数据上的模式识别试验表明，MR-GRU模型准确率优于传统模型。     

Comparison of observed ionospheric vertical TEC over the sea in Indian region with IRI-2016 model

The vertical ionospheric TEC values obtained from GAGAN grid based ionospheric delay correction values over the sea in the Indian equatorial region have been compared with the corresponding values derived from the International Reference Ionosphere model, IRI-2016. The objective of this work is to study the deviation of the vertical TEC derived from the IRI model from ground truths over the sea for different conditions. This will serve the basic intention of assessing the candidature of the IRI model as an alternative ionospheric correction model in navigation receivers in terms of accuracy. We have chosen different solar activity periods, seasons, geomagnetic conditions, locations etc. for our comparison and analysis. The TEC values by the IRI-2016 were compared with the actual measured values for the given conditions and errors were obtained. The measured vertical TEC values at the ionospheric grid points were derived from the GAGAN broadcast ionospheric delay data and used as reference. The IRI model with standard internal functions was used in estimating the TEC at the same ionospheric grid points. The errors in the model derived values are statistically analysed. Broadly, the results show that, for the Indian sector over the sea, the IRI model performs better on quiet days in off equatorial regions, particularly in the northern region. The overall performance degrades for other conditions with the model generally underestimating the true TEC values and most severely in the equatorial region. The performance is worst in this region for the disturbed days of the equinoctial period. The comparison study is also done with the TEC data measured directly by dual frequency GPS receivers. The results were found to be in general agreement with those obtained by comparing the model with GAGAN broadcast data as reference. This study will be useful in considering the IRI-2016 model for real time estimates of TEC as an alternative to the current parametric model in a satellite navigation receiver in absence of other options.     

基于DNET的空中红外目标抗干扰识别算法

复杂空战背景下针对人工干扰的博弈是红外空空导弹精确探测制导技术发展面临的瓶颈和核心技术。针对人工干扰对空中红外目标产生的遮蔽、黏连、相似等干扰现象，以及目标机动和相对运动造成的形状、尺度、辐射特性剧烈变化等实际问题，提出一种基于信息特征提取的深度卷积神经网络DNET空中红外图像目标抗干扰识别算法。首先，DNET网络对大尺度特征图像采用密集连接模块，在前部通道保存每一层的网络输出，在网络末端引入特征注意力机制，获得每个特征通道的信息特征识别权重。然后，加入多尺度密集连接模块，并与多尺度特征融合检测结合，提高对大尺度变化情况下的目标特征提取能力。实验结果表明，在伴随红外诱饵干扰的实时检测条件下，红外目标由点目标变化为成像目标，直至充满视场的整个过程中，本文抗干扰识别算法的识别精确度、召回率及识别速度分别达到99.36%、96.95%、132 fps，具备识别精确度和召回率高、识别速度快等优点，并具有良好的鲁棒性。     

基于图像识别的机械振动信号特征提取与寿命预测方法研究

滚动轴承作为许多机械设备的关键组件,被广泛应用于机械制造、航空航天等领域,其健康状态直接影响了相应设备的剩余寿命,因此在设备故障预测与健康管理(Prognostics and Health Management, PHM)领域,滚动轴承寿命预测具有很高的研究价值。目前基于数据驱动的轴承寿命预测方法主要利用特征提取并构造健康因子(Health Indicator, HI),然而在这一过程中特征的选择与融合依然依赖于专家先验知识,并且健康因子也很难从复杂的时序数据中进行提取。因此,提出了一种新型的数据驱动寿命预测算法,在特征提取方面,通过连续小波变换(Continuous Wavelet Transform,CWT)将传感器振动信号转换为时频谱图,再通过深度残差网络(Deep residual network, ResNet)结合时空卷积网络(Temporal Convolutional Network, TCN)将时频谱图中的时域频域特征构造成为健康因子,最后完成剩余寿命预测。本研究在PRONOSTIA数据集上与现有的数据驱动算法进行了对比,证明了该算法可以更准确地完成剩余寿命预测。     

面向快速响应产品的星上目标定位及识别方法

针对遥感卫星星上目标的定位及识别，本文面向快速响应应用需求，首先对星上目标定位及识别工作模式进行研究，总结出狭义和广义两种模式，其次重点对像方-物方坐标双向定位算法进行梳理，并针对典型地物目标星上自动识别算法提出研究思路，最后给出算法的应用方法流程。本文研究为快速响应产品星上智能处理提供方法参考。     

Non-stationarity and cross-correlation effects in the MHD solar activity

The analysis of turbulent processes in sunspots and pores which are self-organizing long-lived magnetic structures is a complicated and not yet solved problem. The present work focuses on studying such magneto-hydrodynamic (MHD) formations on the basis of flicker-noise spectroscopy using a new method of multi-parametric analysis. The non-stationarity and cross-correlation effects taking place in solar activity dynamics are considered. The calculated maximum values of non-stationarity factor may become precursors of significant restructuring in solar magnetic activity. The introduced cross-correlation functions enable us to judge synchronization effects between the signals of various solar activity indicators registered simultaneously.     

Latitudinal distribution of solar coronal active regions

We studied the cyclic evolution of the latitudinal distribution of solar coronal active regions based on daily images from SOHO EIT for the period 1995–2017. Fully automated software was used, which included the following steps: initial preparation of images in the data series, normalization of histograms and correction of limb brightening, segmentation of images using threshold intensity values obtained from their histograms, scanning of segmented images in heliographic coordinates and obtaining profiles of latitudinal distribution of coronal active regions for each image of the data series. From the output data, we obtained a temporary change in the latitudinal distribution profiles and the migration of activity centers on the solar disk. From the period of minimum activity to the next minimum in both hemispheres, activity centers begin to migrate from high latitudes towards the equator. At the same time, the general center of activity repeatedly changes the direction of migration. The latitudinal distribution of the so-called presence factor of coronal active regions closely resembles the magnetic butterfly diagram, which proves their direct causal relationships. Variations in the presence factor of coronal active regions are correlated with cyclic variations in the sunspot daily numbers.     

一种虹膜特征提取与匹配方法

人眼虹膜尺寸很小，并且容易受到干扰，如果不能有效地提取出稳定的虹膜特征进行模式匹配，将严重影响虹膜识别的准确性和鲁棒性。本文提出一种稳定的虹膜特征提取与匹配方法，在虹膜注册端通过构建强特征分类器获得增强型特征模板，在虹膜识别端通过多样本特征映射融合提取稳定的特征样本，并根据风险预测自适应地确定分类阈值，然后进行特征匹配与虹膜分类，从而有效提高虹膜识别的准确性和鲁棒性。     

How do interplanetary shock impact angles control the size of the geoeffective magnetosphere?

In this paper, we investigate temporal and spatial magnetosphere response to the impact of interplanetary (IP) shocks with different inclinations and speeds on the Earth’s magnetosphere. A data set with more than 500 IP shocks is used to identify positive sudden impulse (SI⁺) events as expressed by the SuperMAG partial ring current index. The SI⁺ rise time (RT), defined as the time interval between compression onset and maximum SI⁺ signature, is obtained for each event. We use RT and a model suggested by Takeuchi et al. (2002) to calculate the geoeffective magnetospheric distance (GMD) in the shock propagation direction as a function of shock impact angle and speed for each event. GMD is a generalization of the geoeffective magnetosphere length (GML) suggested by Takeuchi et al. (2002), defined from the subsolar point along the X line toward the tail. We estimate statistical GMD and GML values which are then reported for the first time. We also show that, similarly to well-known results for RT, the highest correlation coefficient for the GMD and impact angle is found for shocks with high speeds and small impact angles, and the faster and more frontal the shock, the smaller the GMD. This result indicates that the magnetospheric response depends heavily on shock impact angle. With these results, we argue that the prediction and forecasting of space weather events, such as those caused by coronal mass ejections, will not be accurately accomplished if the disturbances’ angles of impact are not considered as an important parameter within model and observation scheme capabilities.