控制点与检查点数量和比例对水深遥感反演精度的影响分析

Analysis of the influence of the amount and proportion of control points and check points on the accuracy of bathymetry remote sensing inversion

在水深遥感反演半经验模型中,水深控制点和检查点选取是一项非常重要的工作,前者用于建立实测水深值与遥感影像灰度值之间的数量关系,后者用于评价水深反演精度。但前人在相关研究中并未就控制点和检查点的选取数量和比例给出有实验依据的建议,亦未开展其对水深反演精度的影响分析。针对这个问题,作者使用6020个实测水深点和World View-2、GF-1 WFV、Landsat8 OLI 3种多光谱影像,基于三波段水深反演模型开展了81组实验,分析比较了不同数量和比例的控制点与检查点对反演结果的影响,评价过程利用平均绝对误差(Mean Absolute Error,MAE)、平均相对误差(Mean Relative Error,MRE)和反映实测水深与反演水深相关性的决定系数R2等3种参数。结果表明:(1)当控制点数量达到31个时,水深反演精度即趋于稳定;(2)检查点数量在30个时,其评价指标已可以代表模型反演精度;(3)控制点和检查点的数量比例对反演精度并无影响,但控制点选取过多则会产生反演精度降低的现象,采用人工选取检查点时剔除浅水区的异常点即可有效避免这一问题。

Selection of control points and check points is an important work in semi-empirical model for remote sensing depth inversion. Control points are used to establish the mathematical relationship between the measured water depth and the gray values of remote sensing image. Check points are used to evaluate the accuracy. However, there are no experimental suggestions on the amount and proportion of control points and check points given by previous studies, nor water depth retrieval accuracy analyses were made. In this paper, 6020 measured depths and World View-2, GF-1 WFV and Landsat8 OLI multispectral images are applied to carry out 81 experiments based on 3-band water depth inversion model. Mean Absolute Errors（MAE）, Mean Relative Errors（MRE） and R^2 are analyzed, and the results show that：（1） The inversion accuracy tends to be steady when the amount of control points comes to 31.（2） The indexes can represent the accuracy of inversion when the amount of check points comes to 30.（3） The proportion of control points and check points has no effect on the accuracy of inversion, but the accuracy will be lowered if too many control points were selected. This phenomenon can be avoided by eliminating the exceptional points manually.