共查询到20条相似文献，搜索用时 46 毫秒
1.
时空Kalman滤波可对变形监测数据进行时空滤波去噪、数据插补和变形预测,本文利用时空Kalman滤波进行变形分析,从模型原理及试验两方面比较分析了Kriged Kalman filter(KKF)、space time Kalman filter(STKF)和spatio-temporal mixed effects(STME) 3种典型时空Kalman滤波模型的性能和适用性。结果表明:3种时空Kalman滤波模型均基于空间基函数及动力学模型组合形式描述时空数据的时空相关性,其主要差异在于空间变异的描述形式不同、空间基函数和状态转移矩阵构造过程不同及模型降维方法不同。在适用性方面,KKF模型更适合于稀疏测站的变形分析,STKF模型及STME模型更适合于海量测站的变形分析。在变形分析应用效果方面,3种时空Kalman滤波模型均具有较高精度的时空滤波去噪、数据插补和变形预测性能,其滤波结果相对于普通Kalman滤波结果的平均改善率为21.1%,其缺失数据插补结果相对于Hermite时间插值结果的平均改善率为42.4%,其空间预测结果相对于Kriging空间插值结果的平均改善率为65.3%,其对已知测站未来变形的时空预测结果相对于普通Kalman滤波时间预测结果的平均改善率为20.6%,其对非观测站点未来变形的时空预测结果相对于Kalman滤波+Kriging组合模型预测结果的平均改善率为20.5%。 相似文献
2.
针对传统的变形监测建模方法一般针对单一监测点的变形预测模型,未考虑到监测点间相互作用的变形特点,该文分析了变形监测点间的相互关联性,通过相关系数法对监测点进行分类,并将邻近监测点的观测序列值作为和时间因素等同的影响因子应用到建模过程中,利用高斯过程算法进行训练,建立预测模型。为提高高斯过程算法的模型预测精度,应选择适合工程案例最优协方差函数。通过实例分析,比较GM(1,1)、多点灰色预测模型和顾及邻近点变形因素的高斯过程等3种模型在基坑围岩、滑坡等变形监测数据处理中的预测精度,表明该文算法考虑到监测点间的变形关联性,充分利用高斯过程在针对小样本、非线性数据建模时的高自适应性等优点,具有较高的预测精度。 相似文献
3.
针对动态场景中运动路标点严重影响传统视觉自主定位算法精度,甚至产生定位失效的问题,提出一种顾及动态路标点的稳健高斯混合模型。在传统图优化视觉定位模型的基础上,增加“运动指数”描述图优化模型中路标点的运动概率,把传统图优化高斯模型增强为高斯混合模型,以约束运动路标点对图优化结果的影响;为增强模型对噪声的稳健性,采用方差膨胀模型约束残差方程;详细推导了该高斯混合模型的期望-最大化求解方法,把该问题转化为经典迭代最小二乘问题进行解算。仿真试验和真实数据试验表明：强动态场景中,提出的算法绝对精度指标和相对精度指标均优于传统优化算法;静态或弱动态场景中,提出的算法仍与传统优化算法定位性能相当。本文方法可有效减小场景中运动路标点对优化结果的影响,更适用于移动机器人的自主定位。 相似文献
4.
天顶对流层延迟(zenith tropospheric delay,ZTD)是影响GPS定位精度的关键因素,为了提高ZTD的预测精度,提出一种基于相空间重构的高斯过程回归预测模型.针对ZTD时间序列的混沌特性,利用国际GNSS服务(International GNSS Service,IGS)站提供的ZTD数据,采用C... 相似文献
5.
提出了3种基于影像重叠区相关系数的视频影像关键帧快速提取算法。该算法首先采用多项式快速、准确地拟合出视频影像相关系数的变化趋势,然后利用相邻关键帧影像重叠区的高相关特性实现关键帧帧位的快速、准确提取,并通过实例与基于重叠度计算关键帧算法的时效性进行对比。结果表明,该算法单次计算相关系数的时间相对于单次重叠度计算时间缩短了近4倍,准确度达到88%,且针对不同拍摄手段下的视频影像,提取的关键帧影像精度较高。同时,还对基于视频提取的关键帧影像进行全景图拼接,结果显示未出现明显场景缺失,且可靠性高,适用性强。 相似文献
6.
7.
针对时间差分载波相位/捷联惯导紧组合系统在非高斯噪声环境工作时,采用高斯混合滤波遇到的混合模型参数估计问题,提出了一种变分贝叶斯学习优化的高斯混合自适应滤波算法。该算法借鉴变分学习理论,准确高效地实现了高斯混合模型参数的自适应估计,进一步精化了滤波算法中的随机模型,能够显著提高估计精度,降低计算负担,改善滤波性能。实验结果表明,相比传统滤波算法,该算法的估计精度得到了进一步改善,运算耗时仅与拓展卡尔曼滤波相当。 相似文献
8.
Alexander Stark Jürgen Oberst Frank Preusker Steffi Burmeister Gregor Steinbrügge Hauke Hussmann 《Journal of Geodesy》2018,92(9):949-961
We report on recent refinements and the current status for the rotational state models and the reference frames of the planet Mercury. We summarize the performed measurements of Mercury rotation based on terrestrial radar observations as well as data from the Mariner 10 and the MESSENGER missions. Further, we describe the different available definitions of reference systems for Mercury and obtain the corresponding reference frame using data provided by instruments on board MESSENGER. In particular, we discuss the dynamical frame, the principal-axes frame, the ellipsoid frame, as well as the cartographic frame. We also describe the reference frame adopted by the MESSENGER science team for the release of their cartographic products, and we provide expressions for transformations from this frame to the other reference frames. 相似文献
9.
Lunar Laser Ranging (LLR) provides various quantities related to reference frames like Earth orientation parameters, coordinates and velocities of ground stations in the Earth-fixed frame and selenocentric coordinates of the lunar retro-reflectors. This paper presents the recent results from LLR data analysis at the Institut für Erdmessung, Leibniz Universität Hannover, based on all LLR data up to the end of 2016. The estimates of long-periodic nutation coefficients with periods between 13.6 days and 18.6 years are obtained with an accuracy in the order of 0.05–0.7 milliarcseconds (mas). Estimations of the Earth rotation phase \(\Delta \)UT are accurate at the level of 0.032 ms if more than 14 normal points per night are included. The tie between the dynamical ephemeris frame to the kinematic celestial frame is estimated from pure LLR observations by two angles and their rates with an accuracy of 0.25 and 0.02 mas per year. The estimated station coordinates and velocities are compared to the ITRF2014 solution and the geometry of the retro-reflector network with the DE430 solution. The given accuracies represent 3 times formal errors of the parameter fit. The accuracy for \(\Delta \)UT is based on the standard deviation of the estimates with respect to the reference C04 solution. 相似文献
10.
11.
12.
Reconstruction of 3D models through integrating vertical and oblique imagery has been studied extensively. For a 3D reconstruction, object point cloud coordinates could be calculated using direct georeferencing (DG) obtained from the direct orientation data of a GPS/INS system. This paper implemented DG approaches for vertical and oblique imagery in the earth centered earth fixed frame (e-frame), local tangent frame (l-frame), and map projection frame (p-frame), respectively. In the p-frame, the earth curvature correction formulas were derived through naturalizing oblique imagery to vertical imagery to achieve a high positioning precision. Five basic stereo-pair models for vertical and oblique imagery were simulated to verify the positioning accuracy of different frames. Simulation experiments showed that DG in the e-frame and l-frame of these five scenarios were rigorous, and no systematic errors were imported by the DG model as these frames are Cartesian. DG in the p-frame has obvious systematic errors which are aroused by the earth curvature and projection deformation unconformity in the vertical and horizontal directions. These errors, however, can be compensated effectively through correcting image coordinates of the oblique imagery by extending the standard image coordinate correction approach and the exterior orientation (EO) height term. After the correction, the absolute positioning error is lower than 1/20 GSD for simulation test-1. In the p-frame, the process is straightforward, and it is convenient for producing maps. For high accuracy DG, though, it is recommended to adopt e-frame or l-frame options. 相似文献
13.
滑坡通常发生突然,破坏力巨大,经常造成重大生命安全事故和财产损失。高可靠性、高精度及具有抗差性能的滑坡形变监测预测手段和方法对于国家防灾减灾需求具有切实意义。InSAR技术是一种能够全天时和全天候观测获取高空间分辨率和宽覆盖率影像,高灵敏性捕捉时空维动态变化的监测手段,然而目前应用InSAR时序影像对滑坡区进行滑坡预测的工作仅是凤毛麟角。基于时序InSAR观测结果,本文提出了一种能够有效解决中短期滑坡预测问题的深度学习滑坡预测方法。在三峡新铺滑坡区应用N-BEATS网络模型和Sentinel-1 SAR数据进行形变预测,以均方根误差1.1 mm的预测精度完成了滑坡预测工作,并对预测结果进行了数据结构影响的规律性分析、传统方法效果对比、抗差性评估及置信区间估计等多方位的剖析,结果显示出了其高精度、高可靠性及具有一定抗差能力的突出优势。 相似文献
14.
Development of a Coordinate Transformation method for direct georeferencing in map projection frames
This paper develops a novel Coordinate Transformation method (CT-method), with which the orientation angles (roll, pitch, heading) of the local tangent frame of the GPS/INS system are transformed into those (omega, phi, kappa) of the map projection frame for direct georeferencing (DG). Especially, the orientation angles in the map projection frame were derived from a sequence of coordinate transformations. The effectiveness of orientation angles transformation was verified through comparing with DG results obtained from conventional methods (Legat method1 and POSPac method2) using empirical data. Moreover, the CT-method was also validated with simulated data. One advantage of the proposed method is that the orientation angles can be acquired simultaneously while calculating position elements of exterior orientation (EO) parameters and auxiliary points coordinates by coordinate transformation.These three methods were demonstrated and compared using empirical data. Empirical results show that the CT-method is both as sound and effective as Legat method. Compared with POSPac method, the CT-method is more suitable for calculating EO parameters for DG in map projection frames. DG accuracy of the CT-method and Legat method are at the same level. DG results of all these three methods have systematic errors in height due to inconsistent length projection distortion in the vertical and horizontal components, and these errors can be significantly reduced using the EO height correction technique in Legat’s approach. Similar to the results obtained with empirical data, the effectiveness of the CT-method was also proved with simulated data. 相似文献
15.
时空聚类分析是对时空大数据进行利用的一种有效手段。本文提出了一种分布式增量大数据聚类分析方法,利用分布增量机制不但可以减少重复计算和迁移拷贝次数,而且可以持续对聚类结果进行修正,能够在保持聚类准确性的条件下提升整体运算效率。而聚类算法本身通过数据聚集趋势预分析、聚类算法和结果评价3个步骤,构建了一体化时空邻域,在时间和空间维度保证了聚类结果的准确性。经过试验证明该方法可以实现时空大数据的快速高效信息挖掘。 相似文献
16.
针对现有可降水量预报模型存在预报精度不高等问题,该文提出采用方差分量估计自适应卡尔曼滤波对可降水量数据进行预处理,用以提高径向基神经网络预测模型的预测精度,从而形成高精度预报模型。通过比较不同基站不同时间的数据,分析使用方法的预报精度。实验结果表明:将预测模型应用于全国7个测站进行实验,预测相对精度的平均值可达95%以上,预报残差在10-5左右,残差值小于0.001的占90%以上。在影响因素方面,使用较短时间作为模型原始数据进行预测会得到较好的预测结果。实验证明本预测方法在预报大气可降水量值方面具有较高的精度。 相似文献
17.
针对多种分布形式混合的观测数据,建立了p范混合模型,考虑到模型中混合数属于不完全数据,引入期望最大化(expectation-maximum, EM)算法,对该混合模型的参数进行估计,详细推导了p范混合模型参数估计的迭代公式,并给出了相应的迭代步骤。采用混合高斯分布数据、拉普拉斯分布与高斯分布混合数据及实测GPS观测值残差数据,验证了公式的正确性和适应性。算例结果表明,与单一概率分布相比,p范混合模型能够准确反映数据分布的实际情况,同时利用EM算法估计的模型参数具有较高的精度。 相似文献
18.
Albert K. Chong 《The Photogrammetric Record》2004,19(108):296-310
Generally, a low-cost image-based motion study system consists of a set of two or more video imaging cameras and a set of object space control targets. The control targets are required to provide for the computation of exterior parameters of the video frames of objects in motion. Subsequently, the computed exterior orientation parameters are used to compute the position of the motion targets. In general, the accuracy of the motion data is dependent largely on the accuracy of the 3D coordinates of the control targets, the computed camera and lens parameters and the frame rate of the camera. Obviously, it is difficult to improve the frame rate of a low-cost camera; however, the other factors may be optimised analytically. Optimising the accuracy of the control targets is a straightforward process and is discussed briefly in the paper. Optimising the computed camera and lens parameters was the main focus of the research. Consequently, the paper provides the detail of the developed optimising technique. The results show that an optimal principal distance and other lens parameters can be determined by analysing the error of a set of highly accurate object distances. The evaluation shows that the accuracy of the video motion study system can be improved by as much as six times or a reduction of scaling error from 1·06 to 1·01. 相似文献
19.
假定原始的时空数据由较大尺度下的全局趋势项与较小尺度下的局部偏差项构成,提出了融合空间尺度特性的时空序列建模方法。首先,将原始数据转换为较大尺度下的数据,此部分反映原始数据的趋势特征。然后,将趋势部分剔除,剩余部分反映原始数据的偏差特征。最后,用灰色系统模型和BP神经网络模型分别对趋势项和偏差项建模,它们的组合预测结果即为原始时空序列预测值。采用该方法对实际的年降水数据以及日平均PM2.5浓度数据进行预测建模分析,实验结果表明:融合空间尺度特性的时空序列预测模型可以用于多空间尺度预测,并且预测精度优于不考虑空间尺度特性的建模方法。 相似文献
20.
大速率、不均匀的地面沉降已经威胁到人类的生产活动,高精度的沉降预测结果对于地质灾害的精准防控具有重要意义。为掌握地面沉降的演化规律,利用现场观测数据或InSAR数据开展了多项预测研究。然而,由于空间异质性的存在,大范围地面沉降的准确预测仍然是一项挑战。在这项研究中,从数据驱动的角度提出了一种顾及空间异质性的大范围地面沉降时空预测方法STLSTM（Spatio-temporal Long Short-Term Memory）。首先,通过聚类识别地理空间中的均质子区;然后,在每个子区中,一个特别的长短期记忆LSTM（Long Short-Term Memory）网络被用来捕捉局部位置的非线性特征;最后,利用预训练的网络对未来时刻的地面沉降进行定量预测。在实验部分,哨兵1号影像数据被用来比较STLSTM与其他8种基准方法的性能,利用空间统计指标分析了模型的有效性。结果表明,STLSTM在152 s内达到了最高的预测精度（71.4%）,且能够有效弱化空间异质性对大区域沉降预测任务的影响。总之,这项研究将空间异质性处理策略融合到深度学习模型中,实现了高精度、高时效的大范围地面沉降时空预测。 相似文献