基于EfficientNet的星系形态分类研究

星系的结构和形态能够反映星系自身的物理性质,其形态的分类是后续分析研究的一个重要环节.EfficientNet模型使用复合系数对深度网络模型的深度、宽度、输入图像分辨率进行更加结构化的统一缩放,是一种新的深度网络优化扩展方法.将该模型应用于星系数据形态的分类研究中,结果表明基于EfficientNetB5模型的平均准确率、精确率、召回率以及F1分数(精确率与召回率的调和平均数)都在96.6%以上,与残差网络(Residual network, ResNet)中ResNet-26模型的分类结果相比有较大的提升.实验结果证明EfficientNet的深度网络优化扩展方法可行且有效,可应用于星系的形态分类.     

Preprocessing of gravity gradients at the GOCE high-level processing facility

One of the products derived from the gravity field and steady-state ocean circulation explorer (GOCE) observations are the gravity gradients. These gravity gradients are provided in the gradiometer reference frame (GRF) and are calibrated in-flight using satellite shaking and star sensor data. To use these gravity gradients for application in Earth scienes and gravity field analysis, additional preprocessing needs to be done, including corrections for temporal gravity field signals to isolate the static gravity field part, screening for outliers, calibration by comparison with existing external gravity field information and error assessment. The temporal gravity gradient corrections consist of tidal and nontidal corrections. These are all generally below the gravity gradient error level, which is predicted to show a 1/f behaviour for low frequencies. In the outlier detection, the 1/f error is compensated for by subtracting a local median from the data, while the data error is assessed using the median absolute deviation. The local median acts as a high-pass filter and it is robust as is the median absolute deviation. Three different methods have been implemented for the calibration of the gravity gradients. All three methods use a high-pass filter to compensate for the 1/f gravity gradient error. The baseline method uses state-of-the-art global gravity field models and the most accurate results are obtained if star sensor misalignments are estimated along with the calibration parameters. A second calibration method uses GOCE GPS data to estimate a low-degree gravity field model as well as gravity gradient scale factors. Both methods allow to estimate gravity gradient scale factors down to the 10⁻³ level. The third calibration method uses high accurate terrestrial gravity data in selected regions to validate the gravity gradient scale factors, focussing on the measurement band. Gravity gradient scale factors may be estimated down to the 10⁻² level with this method.     

嫦娥一号IIM数据应用处理流程分析

对嫦娥一号干涉成像光谱仪（IIM）数据的特点进行了分析,并就存在的一些问题提出了解决方案,制定了IIM数据应用处理流程,为该数据的正确使用提供方法参考。研究结果表明,在空间域传感器左侧响应偏低,右侧响应偏高;在波谱域长波段响应存在较大偏差。经过绝对定标和辐射畸变校正后的反射率与地基望远镜光谱匹配良好,可以用于应用研究。利用校正后的数据对Aristarchus地区岩石类型开展初步研究的结果表明,该地区在纵向和横向上都存在岩性的多样性。校正后的图像不仅提高了分类精度,还被识别出撞击坑可能存在的滑坡。嫦娥一号IIM能够在全球、区域和局部尺度上以较高的空间分辨率和光谱分辨率获取月表元素和矿物成分信息,有助于深化对月球形成和演化的认识。     

建筑物沉降观测技术方法探讨

随着城市建设的迅猛发展,城市规模越来越大,建筑物越来越高,由此带来的安全隐患越来越大,由于高层建筑一般采用桩基基础,且荷载较大,对高层建筑的内部基础和设备的位置关系有很高的精度要求,其施工安装阶段将给高层建筑本身及周边建筑群体带来复杂的形变影响。为了保障建筑物施工和运营的安全,就必须对其进行沉降观测。本文结合笔者多年的实际工作经验,并结合具体的实例,对高层建筑物沉降原因、沉降观测前的准备工作及沉降观测的实施过程和观测数据的处理方法等进行了论述,最后指出如何通过沉降观测技术准确地反映出高层建筑物的沉降情况。     

大型油罐沉降监测数据处理软件的实现

介绍了针对某批大型油罐的沉降监测数据处理软件制作过程,采用了Excel电子表格软件在VBA可视化编程环境下完成,实现了工程测量数据的及时处理分析。     

基于惯导/天文高精度定姿方法

研究了利用惯性导航系统与天文导航系统进行组合定姿的方法;构造了惯导/天文组合导航系统的状态方程和量测方程;采用卡尔曼滤波技术设计了惯导/天文组合定姿算法。设计飞行器的典型飞行轨迹,通过数学仿真对该算法的有效性进行了验证。结果表明,天文导航能够有效地补偿惯性导航因陀螺漂移带来的误差,非常适用于飞行器的高精度定姿。     

利用GPU实现基于物理模型的流体运动仿真

纳维-斯托克斯方程是描述流体运动的基础物理模型。首先简要介绍了在计算机上实现该模型的数值解算方法。在流体运动的计算机视觉仿真中,密度渲染才是最终结果,因此,又阐述了流体的密度场与速度场的关系,并且给出了流体仿真的基本流程。由于可编程图形处理器(GPU)具有并行性和速度快等特点,流体模拟的解算和渲染在图形处理器上运行更加高效,但是利用GPU编程实现流体仿真必须要实现3个核心转换,最后以支持3维纹理的Direct3D10为例,给出了利用GPU实现流体运动仿真两个简单实例。     

三维城市模型的研究与实践(英文)

The way we interact with spatial data has been changed from 2D map to 3D Virtual Geographic Environment (VGE). Three-dimensional representations of geographic information on a computer are known as VGE, and in particular 3D city models provide an efficient way to integrate massive, heterogenous geospatial information and georeferenced information in urban areas. 3D city modeling (3DCM) is an active research and practice topic in distinct application areas. This paper introduces different modeling paradigms employed in 3D GIS, virtual environment, and AEC/FM. Up-to-date 3DCM technologies are evolving into a data integration and collaborative approach to represent the full spatial coverage of a city, to model both aboveground and underground, outdoor and indoor environments including man-made objects and natural features with 3D geometry, appearance, topology and semantics. Supported by the National Natural Science Foundation of China ( No. 40871212, No. 40671158), the Leading Academic Discipline Project of Shanghai Educational Committee( No.J50104).     

Total electron content processing from GPS observations to facilitate ionospheric modeling

With the increasing global distribution of high rate dual-frequency global positioning system (GPS) receivers, the production of a real-time atmospheric constituent definition, total electron content (TEC), has become a beneficial contributor to the modeling applications used in the assessment of GPS position accuracy and the composition of the ionosphere, plasmasphere, and troposphere. Historically, TEC measurements have been obtained through post processing techniques to produce the quality of data necessary for modeling applications with rigorous error estimate requirements. These procedures necessitated the collection of large volumes of data to address the various abnormalities in the computation of TEC associated with the use of greater data quality controls and source selection while real-time modeling environments must rely on autonomous controls and filtration techniques to prevent the production of erroneous model results. In this paper we present methods for processing TEC in real time, which utilize several procedures including the application of an ionospheric model to automatically perform quality control on the TEC output and the computational techniques used to address receiver multipath, faulty receiver observations, cycle-slips, segmented processing, and receiver calibrations. The resulting TEC measurements are provided with rigorous error estimates validated using the vertical TEC from the Jason satellite mission.

陡倾斜脉状矿体隐伏深度计算

在研究广西 10 0个矿区地质资料基础上 ,建立了陡倾斜脉状矿体隐伏深度计算经验公式 ,从而为隐伏矿床预测、减少矿床勘查风险提供了信息