一种快速有效的大数据区域网平差方法

发布时间：2018-06-22 22:08

本文选题：区域网平差 + 预条件共轭梯度　；参考：《测绘学报》2017年02期

【摘要】：针对摄影测量影像来源多样化、复杂化、大数据化等趋势,传统区域网平差算法在应对当前复杂多变的数据来源,矩阵排列毫无规律的法方程结构以及大数据量带来的高内存需求和低计算效率等问题上,遇到了前所未有的挑战,为了解决上述难题,本文引入了预条件共轭梯度法以及不精确牛顿解法求解区域网平差过程中的法方程,同时使用一种块状法方程系数矩阵压缩存储格式,构建了全新的区域网平差技术流程。本文方法避免了直接对法方程系数矩阵的求逆,压缩了法方程系数矩阵所需的内存空间,使得本文算法比传统算法所需计算机内存空间大幅减少,平差计算速度明显提升,同时保证了计算精度与传统方法相当。初步试验证明,本文方法对4500张影像、近900万像点数据的平差计算在普通电脑上仅需要约15min,且计算精度达到子像素级。
[Abstract]:In view of the trend of photogrammetric image source diversification, complexity and large data, the traditional area network adjustment algorithm is dealing with the current complex and changeable data sources. In order to solve the above problems, the irregular normal equation structure of matrix arrangement and the problems of high memory requirement and low computing efficiency caused by large amount of data have met unprecedented challenges. In this paper, the preconditioned conjugate gradient method and the inexact Newton method are introduced to solve the normal equations in the adjustment process of the regional network. At the same time, a new technique flow of the regional network adjustment is constructed by using a block method equation coefficient matrix compressed storage format. This method avoids the direct inversion of the normal equation coefficient matrix, compacts the memory space required by the normal equation coefficient matrix, and greatly reduces the computer memory space required by the traditional algorithm, and improves the speed of adjustment calculation. At the same time, the calculation accuracy is equal to that of the traditional method. The preliminary test shows that the adjustment of 4500 images and nearly 9 million image points in this paper takes only about 15 minutes on a common computer, and the calculation accuracy reaches sub-pixel level.
【作者单位】：中国地质大学(武汉)国家地理信息系统工程技术研究中心;武汉大学遥感信息工程学院;北京中测智绘有限责任公司;
【基金】：国家自然科学基金(41601502;41571434) 中国博士后科学基金面上项目(2015M572224) 中央高校基本科研业务费专项资金(CUG160838)~~
【分类号】：P207;P23

【相似文献】