野值预剔除方法的改进

在初轨有较大误差时，传统的光学资料野值预剔除方法——截断最小二乘L估计法不能有效地剔除野值．从门限和模型两方面改进野值预处理方法，即改用统计中误差定门限．将残差的线性模型改为2次曲线模型，结果为：对于模拟产生的所有卫星弧段，在初轨有10^-4（半长轴α为其5％，以下不再说明）误差情况下，野值比例都能降低到10％以下．

An Improvement Method for Data Preprocess

When initial orbit errors are too big, the traditional optical data preprocess method-interceptive least-square L-estimation method can not delete outliers effectively. This method is improved by thresholds and model: the thresholds are determined by statistical mean error; the linear model is changed to quadratic curve model. The conclusions are: for all satellite passes which are simulated, when the initial orbit error is 10-4 (semi-major is 5 percent of it, which don't explain in the following) ,the proportion of outliers is reduced less than 10% in all cases.