李德仁. 自检校平差中过度参数化的克服[J]. 武汉大学学报 ( 信息科学版), 1986, 11(3): 95-104.
引用本文: 李德仁. 自检校平差中过度参数化的克服[J]. 武汉大学学报 ( 信息科学版), 1986, 11(3): 95-104.
Li Deren. The Overcoming of the Overparametrization in Self-Calibrating Adjustment[J]. Geomatics and Information Science of Wuhan University, 1986, 11(3): 95-104.
Citation: Li Deren. The Overcoming of the Overparametrization in Self-Calibrating Adjustment[J]. Geomatics and Information Science of Wuhan University, 1986, 11(3): 95-104.

自检校平差中过度参数化的克服

The Overcoming of the Overparametrization in Self-Calibrating Adjustment

  • 摘要: 带附加参数的自检校平差是当今解析空中三角测量中补偿系统误差的最有效方法。可惜的是,当采用一个十分庞大的附加参数组时,在某些几何条件下将导致很坏的法方程系状态,从而使测求的平差结果很差。本文研究和试验了几种克服过度参数化的方法。它们不是用统计分析方法来消除某些附加参数的,因为这会引起实际计算的困难。本试验研究表明:——如果采用飞行方向交叉的双区域,用自由附加参数平差,附加参数的可测定性和系统的可靠性均是可接受的;——如果附加参数作为带权观测值,而且按信噪比确定其权,在许多情况下过度参数化将是可避免的。——如果所有的未知数按极大验后估计原理或岭估计原理处理成带权观测值,並给予一个小权(例如P=0.005~0.001),则在所试验的任何情况下均可克服过度参数化。

     

    Abstract: The selfcalibrating adjustment with additional parameters is the most efficient method for the compensation of systematic errors in analytical aerotriangulation today. Unfortunately, introducing a comprehensive set of additional parameters can produce a bad condition for normal equations under certain geometry, so that the adjusted results are poorly determined. In this paper some ways for overcoming the overparametization are investigated which do not eliminate some additional parameters by means of a statistic analysis, which leads to computational diffiiculties in practice. The investigation indicates: ——If a twofold block with cross flight direction is used, and the adjustment is performed with free additional parameters, the determinability of additional parameters and the reliability of the system will both be acceptable. ——If the additional parameters are considered as weighted observations where the weights are determined by the signal-to-noise ratio. the overparametrization can be in many cases avoided. ——If all the unknowns by the principle of the Maximum-Posteriori-Estimation or Ridge-Estimation are considered as weighted observations with a small weight (e.g.p=0.005—0.001), the overdarammetrization will be able to be overcome in any case of our experiments efficiently.

     

/

返回文章
返回