遮挡条件下的鲁棒表情识别方法

提出了一种对面部遮挡具有鲁棒性的表情识别方法.首先,基于鲁棒主成分分析(RPCA,Robust Principal Component Analysis)对待识别人脸进行重构,并对重构人脸和待识别人脸的差值图像进行显著性检测得到面部遮挡区域;其次,将待识别人脸的遮挡区域由RPCA重构人脸的相应区域进行替换,并由权值更新的AdaBoost分类器对遮挡区域重构后的人脸进行表情识别.在BHU(Beihang University)人脸表情数据库和日本女性表情数据库上进行了各种遮挡情况下的表情识别实验,获得了比AdaBoost方法更好的识别结果,说明基于RPCA和AdaBoost的表情识别方法对多种面部遮挡具有较强的鲁棒性.     

Variations of Thermosphere Density Based on Joint Analysis of In-Situ Measurement Data From Shenzhou Spacecrafts

Based on the measurements made by Atmospheric Density Detectors (ADDs) onboard Chinese spacecraft Shenzhou 2-4, the variations of thermosphere density are revealed. During the quiet period, the density at spacecraft altitude of 330～410km exhibited a dominant diurnal variation, with high value on dayside and low value on nightside. The ratio of the diurnal maximum density to the minimum ranged from 1.7 to 2.0. The ratio shows a positive correlation with the level of solar activity and a negative correlation with the level of geomagnetic activity. When a geomagnetic disturbance comes, the atmospheric density at the altitude of 330～410km displayed a global enhancement. For a strong geomagnetic disturbance, the atmospheric density increased by about 56%, and reached its maximum about 6～7 hours after the geomagnetic disturbance peak. The density asymmetry was also observed both in the southern and northern hemisphere during the geomagnetic disturbance peak.      

太阳周期活动对低高度内辐射带高能质子的影响

利用NOAA-15卫星1998年到2011年近13年的高能质子全向通量观测资料, 分析了一个太阳活动周内, 低高度内辐射带高能质子通量的分布变化特性及其物理原因, 比较了观测结果与AP8模型的不同. 研究表明, 低高度内辐射带高能质子通量与太阳活动水平的反相关关系与磁壳参数L值及磁场B值有关; L值越低, B值越大的空间点, 其高能质子通量与太阳活动水平的反向相关性越明显. 高能质子通量随太阳活动水平的变化存在明显滞后现象, L值越高、 B值越小的空间点, 滞后现象就越明显, 滞后严重时可以达到一年左右的时间; 这种滞后现象反映出低高度内辐射带高能质子的源与损失达到平衡是一个中长期过程. 通过与AP8模型计算结果的比较分析可以看出, 利用AP8模型时, 仅考虑地磁场长期变化对质子通量的影响可能会夸大低高度内辐射带局部高能质子通量的增强.      

A brief overview of the School of Aerospace Engineering of Tsinghua University

This article provides a brief overview of the teaching and research at the School of Aerospace Engineering (SAE) to celebrate the 80th anniversary of the establishment of aeronautics as a discipline at Tsinghua University. The evolution of the school, undergraduate/graduate students and faculty members, and research activities and achievements have been described. The research input including research funding and research projects are summarized, showing a diversity of funding sources and a significant growth in either sum total or spending per researcher. The achievements including publications and inter/national academic awards are also introduced. It can be seen that the level of academic publications has been growing over the past decades. In addition, four representative research achievements have been briefly described to show the scientific contributions of the school.     

Geoeffectiveness of interplanetary shocks controlled by impact angles: A review

The high variability of the Sun’s magnetic field is responsible for the generation of perturbations that propagate throughout the heliosphere. Such disturbances often drive interplanetary shocks in front of their leading regions. Strong shocks transfer momentum and energy into the solar wind ahead of them which in turn enhance the solar wind interaction with magnetic fields in its way. Shocks then eventually strike the Earth’s magnetosphere and trigger a myriad of geomagnetic effects observed not only by spacecraft in space, but also by magnetometers on the ground. Recently, it has been revealed that shocks can show different geoeffectiveness depending closely on the angle of impact. Generally, frontal shocks are more geoeffective than inclined shocks, even if the former are comparatively weaker than the latter. This review is focused on results obtained from modeling and experimental efforts in the last 15?years. Some theoretical and observational background are also provided.     

一种多运动模式下自适应阈值零速修正算法

传统基于微机电惯性测量单元（MEMS-IMU）的惯性导航系统（INS）引入零速修正（ZUPT）算法校正器件的累积误差。但由于ZUPT算法零速判定阈值为固定值,只适合单一运动模式,当室内行人运动轨迹包含多种运动模式时,定位精度下降。对此,提出了一种多运动模式下自适应阈值ZUPT算法。分析了室内行人包括静止、走、跑、上楼和下楼5种运动模式零速判定阈值的选取,实现了利用随机森林（RF）算法对5种运动模式的分类识别,并根据识别结果对ZUPT算法零速判定阈值进行自适应调整。为了验证本文算法的可行性和有效性,利用MATLAB软件平台对实测数据进行处理,并与传统定位算法进行了比较。3组实验结果表明,当室内行人运动轨迹包括多种运动模式时,相比传统固定阈值的ZUPT算法,引入自适应调整阈值的ZUPT算法可使定位算法的定位精度提高73.83%。      

一种基于深度学习的交互式电话号码识别方法

物流、保险和中介服务等行业需要频繁地拨打电话,而人工拨打电话效率较低,高效的电话号码识别技术具有重要的应用价值。传统的印刷体数字识别方法存在人工设计特征过程复杂、识别字体单一等不足,难以满足实际应用需求。本文提出了一种基于深度学习的交互式的电话号码识别方法,通过鼠标双击图像中的电话号码,自动截取出包含此号码的目标区域,并进行灰度化、二值化、目标区域定位、字符分割和图片补白等预处理操作,在此基础上利用改进的LeNet-5卷积神经网络（CNN）自动学习图像特征,支持多种字体、字形和字号的印刷体数字识别,并利用交互式识别和内存池等方法提高识别速度。实验结果表明,单一字符的识别率为99.86%,整个号码的识别率为99.50%,整个号码平均识别时间为91 ms。本文方法识别精度高、识别速度快,具有较为广泛的应用前景。      

基于改进遗传算法的移动目标成像侦测任务规划问题研究

基于先验信息调用成像侦察卫星监控陆地或海洋移动目标动态信息是卫星成像侦察面临的新 课题。在已知移动目标位置等先验信息基础上,动态构造目标可变潜在区域及其运动预测模 型,利用STK辅助构造其候选成像观测活动集合;在此集合及目标运动预测模型基础上 对动态可变区域成像卫星调度问题进行建模,并设计了一种基于模拟退火算法及遗传算法的 改进遗传算法对问题进行求解,得到近最优的移动目标成像侦测方案。最后通过实例及算法 对比验证了规划模型及算法对解决该类问题的合理性和有效性。
     

“国际空间站”建造十年经验初探

就＂国际空间站＂边建造、边应用十年来所获得的经验与教训,从系统设备、居住环境、物资补给、作息制度、健康保障、人的作用、出舱活动及国际合作8个方面进行了简要的论述,对后来的空间站建设及未来的长期载人航天飞行具有重要的参考价值。