Outlier Detection Method And Application For EVI Model

In the processing of measured data, Least Square method has become the most commonly used data processing methods for Surveying and mapping workers, because of the many advantages of the least square method in parameter estimation and calculation.But many advantages of this method will be broken when there is model error in the measured data(this paper mainly refers to outliers), so that the result of adjustment is not reliable. In order to ensure the reliability and accuracy of adjustment results, it is very important to study how to find and eliminate the outliers in observation data. At present,the outlier detection methods are mainly divided into two categories, the mean shift model and the variance covariance expansion model. However, most of the outlier detection methods are based on the least square method, considering that there are no random errors or outliers in the coefficient matrix, so it can effectively find the outliers in the observation vector, this is a very ideal situation and this situation does not exist in practice. For example, in the linear fitting, coordinate transformation and the adjustment of the net, the coefficient matrix is contained the measurements, so that the coefficient matrix is very likely to have random errors or outliers. So, it is very necessary to study how to find the outliers in EIV models when there are random errors in both the coefficient matrix and observation vector.Through the reference and study of the existing methods of outlier detection, this paper focuses on the following issues:(1) In this paper, several outlier detection methods of mean shift mode are described and discussed based on Least Square method, including: Data Snooping, Partly least square method, QUAD, LEGE. And the numerical results show that the estimated values of outliers calculated by the above four methods are equal in the case of independent accuracy and independent ranging accuracy.(2) This paper studies the outlier detection methods for EIV model that considering the coefficient matrix and observation vector exist random error. The firstoutlier detection method is established base on WTLS model. Through the deformation of the WTLS function model, and transforming it to standard least squares form, then the data snooping and outlier detection based on partly LS are established; the second outlier detection method is established based on Partial-EIV model. Through the linearization of Partial-EIV model, and transforming it to standard least squares form, then the data snooping and outlier detection based on partly LS are established.(3) For the problem of low computational efficiency of Partial-EIV model, the new algorithms for Partial-EIV model is applied to data snooping that based on Partial-EIV model. The simulation results show that the proposed method can effectively improve the computation efficiency of outlier detection. And for the problem of outliers in the seven parameter transformation model, the data detection method based on Partial-EIV model is applied to the calculation of parameters, and the simulation results show that the method can effectively find the outliers in the process of parameter calculation, so as to ensure the accuracy and reliability of the results.